2013 IEEE Information Theory Workshopitw2013.tsc.uc3m.es/sites/itw2013.tsc.uc3m.es/files/... ·...

Technical and Social Programs & Book of Abstracts

2013 IEEE Information Theory Workshop

Pantone© 137

Tintas planas

Cuatricromía

Pantone© 2738

C=0 M=50 Y=100 K=0

C=100 M=100 Y=0 K=0

Seville, Spain9–13 September 2013

SPONSORS

CONTRIBUTORS

The IEEE Information Theory Society would like to recognize the generous contribution of Universidad CarlosIII de Madrid and of Universidad de Sevilla and in particular its Vicerrectorado de Investigacion and its EscuelaTecnica Superior de Ingenierıa (ETSI):

COLLABORATORS

The conference organization has been supported by the collaboration of University of California San Diego,Northwestern University, Universitat Pompeu Fabra, and Centre of Studies and Technical Research of Gipuzkoa:

ITW 2013 9–13 September 2013, Seville, Spain

WELCOME

¡Bienvenidos to the IEEE Information Theory Workshop 2013!

This year ITW is taking place in Seville, a UNESCO world heritage city for “incorporating vestiges of Islamicculture, centuries of ecclesiastical power, royal sovereignty and the trading power that Spain acquired through itscolonies in the New World.” The conference is organized by the IEEE Information Theory Society and it willtake place at the School of Engineering of the University of Seville. The workshop is supported by the EscuelaTecnica Superior de Ingenierıa, Universidad de Sevilla and Universidad Carlos III de Madrid.

The technical program committee consisted of 35 members who reviewed the papers and scheduled the talks. Wereceived 206 submissions, and after a total of 588 reviews, accepted 120 papers, about 58% of submissions. Weare grateful to the TPC members, to the reviewers, and to the tens of individual contributors that will present theirmost recent work at this event.

We have also scheduled several invited sessions and talks, including six invited sessions organized by their chairs,four hour-long plenaries, and following last year’s ITW, eight half-hour plenaries, affectionately termed “ple-naritas”. We are very pleased to welcome our distinguished invited speakers: Venkat Anantharam, Suhas Dig-gavi, Ernest Fraenkel, Laszlo Gyorfi, Michael Kearns, Amos Lapidoth, Gabor Lugosi, Amin Shokrollahi, NaftaliTishby, Rudiger Urbanke and Martin Wainwright. And last, but not least, we are also looking forward to Mon-day’s reception, which will feature Senor Verdu talking on Don Claude Shannon.

To keep a more leisurely schedule, we organized the talks in three parallel sessions. This allows for longer breaksand ending early on Wednesday and Friday.

We hope you will find the program instructive and interesting.

Enjoy your stay in Seville and the workshop!

The organizers.

3


ORGANIZING COMMITTEE

General Co-Chairs TPC Co-ChairsJuan Jose Murillo-Fuentes Albert Guillen i Fabregas

Fernando Perez Cruz Michael HonigJavier Payan-Somet Alon Orlitsky

Pedro Crespo

Local Arrangements TreasurerLuis Salamanca Matilde Sanchez FernandezPablo M. OlmosRafael Boloix

Eva Arias Publications ChairMelanie F. Pradier Alfonso MartinezCamilo G. Taborda

TECHNICAL PROGRAM COMMITTEE

Albert Guillen i Fabregas Stark Draper Stefan Moser(co-chair) Uri Erez Pierre Moulin

Michael Honig Andrea Goldsmith Prakash Narayan(co-chair) Vivek Goyal Urs Niesen

Alon Orlitsky Dongning Guo Dani Palomar(co-chair) Bruce Hajek Haim Permuter

Stephen Hanly Henry PfisterKhaled Abdel-Ghaffar Rolf Johannesson Max Raginsky

Salman Avestimehr Young-Han Kim Yossi SteinbergMathieu Bloch Sanjeev Kulkarni Sekhar Tatikonda

Paul Cuff Nick Laneman David TseZoran Cvetkovic Angel Lozano Tsachy Weissman

Alex Dimakis Matthew McKay Rebecca WillettLara Dolecek Olgica Milenkovic Raymond Yeung

4


CONFERENCE ACTIVITIES: OVERVIEW

Social Program

The Social Program takes place at different locations in the city centre of Seville.

Monday Evening

Claude Shannon (an edutainment presentation by Sergio Verdu)

Location: Paraninfo, Universidad de Sevilla (C/ San Fernando 4)

Time: 18:45

Welcome Reception

Location: Hotel Los Seises (C/ Segovias, 6)

Time: 20:30

Tuesday Evening

Guided Tour of Seville’s Alcazar Palace

Meeting Point: Real Alcazar, Puerta del Leon entrance

Time: 20:45

Thursday Evening

Conference Banquet

Location: Restaurante Abades (C/ Betis, 69)

Time: 20:30

Technical Program

The Technical Program takes place at the Escuela Tecnica Superior de Ingenierıa (ETSI, School of Engineering).ETSI is located on the Isla de La Cartuja, in one of the buildings used in Seville’s World Fair in 1992. Walking toor from the city centre takes around 30 minutes; the most convenient river bridge is Puente de la Barqueta.

Sessions are scheduled on Tuesday morning and afternoon (till 18:10), Wednesday morning and afternoon (till15:30), Thursday morning and afternoon (till 18:10), and Friday morning (till 12:50).

Lunch (included in the registration fee) will be served at the Conference Venue premises (ETSI) on Tuesday,Wednesday, and Thursday. Since the conference closes on Friday at 12:50, no lunch is offered on that day.

5


6


CONFERENCE VENUEESCUELA TECNICA SUPERIOR DE INGENIERIA

Ground Level (Plenaries, Coffee Breaks, Lunch)

escuela técnica superior de

INGENIERÍA

Ground Floor - Planta Baja (PB)

2 3

Elevator 4ElevatorStairs Stairs

Room:Salón de

Actos

WC WC

WC

WCWC

WC

WC WC

Northern entrances

Main Entrance

1

Coffee Break& Lunch

Coffee Break& Lunch

FloorsPB

E1

P1

E2

PA

7


CONFERENCE VENUE (CONT.)ESCUELA TECNICA SUPERIOR DE INGENIERIA

Floor E2 (Room 3 – Aula 304)


INGENIERÍA

Floor E2 - Entreplanta 2 (E2)

Room 3:Aula 303

& 304

WC

WCWC

WC

WCWC

Elevator 4Elevator1Stairs Stairs

PB

E1

P1

E2

PA

Floors

8


CONFERENCE VENUE (CONT.)ESCUELA TECNICA SUPERIOR DE INGENIERIA

Last Floor (Room 1 – Sala Juan Larraneta & Room 2 – Sala de Grados)


INGENIERÍA

Last Floor - Planta Ático (PA)

Room 1:Sala Juan Larrañeta

Room 2:Sala de Grados

Cafeteria

Stairs 1 4

Elevator

Stairs

Elevator32

WC

PB

E1

P1

E2

PA

Floors

9


10

Sessions

Pantone© 137

Tintas planas

Cuatricromía

Pantone© 2738

C=0 M=50 Y=100 K=0

C=100 M=100 Y=0 K=0


12


Monday, September 9∗∗ Monday’s evening program does not take place in the conference venue, but in the city centre.

18:45–19:45

Claude Shannon (by Sergio Verdu), Location: Paraninfo Univ. de Sevilla (C/ San Fernando 4)

20:30–22:15

Welcome Reception, Location: Hotel Los Seises (C/ Segovias, 6)

Tuesday, September 10

08:40–09:40

From Statistics to Information Theory: Challenges in the Big Data EraMartin J. Wainwright (University of California, Berkeley, USA) (Location: Salon de Actos)

09:50–11:10

TuB1Polar Coding 1

TuB2Networks: Compression and Coding(Organized by Suhas Diggavi andMichelle Effros)

TuB3Secrecy, Privacy and Cryptography

Room 1 – Sala Juan Larraneta Room 2 – Sala de Grados Room 3 – Aula 304

Chair: Joerg Kliewer (New Mexico State University, USA) Chair: Michelle Effros (California Institute of Technology,USA)

Chair: Matthieu Bloch (Georgia Institute of Technology,France)

09:50 Polar Codes with Dynamic Frozen Symbols and TheirDecoding by Directed Search

Peter Trifonov , Vera Miloslavskaya

On Symmetric Multiple Description Coding

Lin Song, Shuo Shao, Jun Chen

Spectrum Bandit Optimization

Marc Lelarge, Alexandre Proutiere, M. Sadegh Talebi

10:10 A Lower Bound on Achievable Rates by Polar Codes withMismatch Polar Decoding

Mine Alsan

Characterising Correlation via Entropy Functions

Satyajit Thakor , Terence H. Chan, Alex Grant

Physical-Layer Cryptography Through Massive MIMO

Thomas Dean, Andrea Goldsmith

10:30 Scaling Exponent of List Decoders with Applications toPolar Codes

Marco Mondelli , S. Hamed Hassani , Rudiger Urbanke

Erasure/List Exponents for Slepian-Wolf Decoding

Neri Merhav

Secrecy & Rate Adaptation for Secure HARQ Protocols

Mael Le Treust , Leszek Szczecinski , Fabrice Labeau

10:50 Polar Coding for Fading Channels

Hongbo Si , Onur Ozan Koyluoglu, Sriram Vishwanath

Operational Extremality of Gaussianity in NetworkCompression, Communication, and Coding

Himanshu Asnani , Ilan Shomorony , Salman Avestimehr ,Tsachy Weissman

Protecting Data Against Unwanted Inferences

Supriyo Chakraborty , Nicolas Bitouze, Mani B. Srivastava,Lara Dolecek

11:30–12:50

TuC1Polar Coding 2

TuC2IT & Coding for Contemporary Video(Organized by Christina Fragouli andEmina Soljanin)

TuC3Information-Theoretic Secrecy


Chair: Olgica Milenkovic (University of Illinois, USA) Chair: Emina Soljanin (Bell Labs, Alcatel - Lucent, USA) Chair: Aylin Yener (Pennsylvania State University, USA)

11:30 Polar Coding for Secret-Key Generation

Remi A. Chou, Matthieu R. Bloch, Emmanuel Abbe

Some Coding and Information Theoretic Problems inContemporary (Video) Content Delivery

Emina Soljanin

Exploiting Common Randomness: a Resource for NetworkSecrecy

Laszlo Czap, Vinod M. Prabhakaran, Suhas Diggavi ,Christina Fragouli

11:50 On Polarization for the Linear Operator Channel

Cesar Brito, Joerg Kliewer

Source Broadcasting over Erasure Channels: DistortionBounds and Code Design

Louis Tan, Yao Li , Ashish Khisti , Emina Soljanin

Secrecy in Cascade Networks

Paul Cuff

12:10 Channel Polarization with Higher Order Memory

Huseyin Afser , Hakan Delic

Impact of Random and Burst Packet Losses on H.264Scalable Video Coding

Siyu Tang, Alface Rondao Alface

Secure Degrees of Freedom of MIMO X-Channels withOutput Feedback and Delayed CSI

Abdellatif Zaidi , Zohaib Hassan Awan, Shlomo (Shitz)Shamai , Luc Vandendorpe

12:30 Network Coding Designs Suited for the Real World: Whatworks, What doesn’t, What’s Promising

Morten V. Pedersen, Daniel E. Lucani , Frank H.P. Fitzek ,Chres Sørensen, Arash Shahbaz Badr

The Secrecy Capacity of a Compound MIMO GaussianChannel

Rafael F. Schaefer , Sergey Loyka

13


Tuesday, September 10 (Cont’d)

14:00–14:30

Polar versus Spatial Coupling: The Good, the Bad, and the UglyRudiger Urbanke (EPFL, Switzerland) (Location: Salon de Actos)

14:30–15:00

Managing Bursty Interference with FeedbackSuhas Diggavi (University of California Los Angeles, USA) (Location: Salon de Actos)

15:10–16:30

TuE1Spatial Coupling

TuE2Ranking and Group Testing

TuE3Multi-terminal Communication


Chair: Krishna Narayanan (Texas A&M University, USA) Chair: Suhas Diggavi (University of California Los Angeles,USA)

Chair: Shlomo (Shitz) Shamai (The Technion, Israel)

15:10 Thresholds of Spatially Coupled Systems via Lyapunov’sMethod

Christian Schlegel , Marat V. Burnashev

Aggregating Rankings with Positional Constraints

Farzad Farnoud (Hassanzadeh), Olgica Milenkovic

Fundamental Limits of Distributed Caching in D2DWireless Networks

Mingyue Ji , Giuseppe Caire, Andreas Molisch

15:30 Displacement Convexity – A Useful Framework for theStudy of Spatially Coupled Codes

Rafah El-Khatib, Nicolas Macris, Rudiger Urbanke

Iterative Similarity Inference via Message Passing in FactorGraphs for Collaborative Filtering

Jun Zou, Arash Einolghozati , Erman Ayday , FaramarzFekri

Analog Index Coding over Block-Fading MISO BroadcastChannels with Feedback

Pablo Piantanida, Mari Kobayashi , Giuseppe Caire

15:50 A Closed-Form Scaling Law for Convolutional LDPCCodes Over the BEC

Pablo M. Olmos, Rudiger Urbanke

Stochastic Threshold Group Testing

Chun Lam Chan, Sheng Cai , Mayank Bakshi , SidharthJaggi , Venkatesh Saligrama

Equivalence of Inner Regions for Broadcast ChannelCoding

Jun Muramatsu

16:10 A Finite Length Performance Analysis of LDPC CodesConstructed by Connecting Spatially Coupled Chains

Pablo M. Olmos, David G. M. Mitchell , Dmitri Truhachev ,Daniel J. Costello, Jr.

Optimal Binary Measurement Matrices for CompressedSensing

Arash Saber Tehrani , Alexandros Dimakis, Giuseppe Caire

A General Framework for Statistically Characterizing theDynamics of MIMO Channels

Francisco Javier Lopez-Martinez, Eduardo Martos-Naya ,Jose Francisco Paris, Andrea Goldsmith

16:50–18:10

TuF1LDPC Codes

TuF2Fundamental Limits of Learning andInference

TuF3Multiple Access Channels

Chair: Christian Schlegel (Dalhousie University, USA) Chair: Marc Lelarge (INRIA and ENS, France) Chair: Dongning Guo (Northwestern University, USA)


16:50 Design of Masking Matrix for QC-LDPC Codes

Yang Liu, Ying Li

An Impossibility Result for High Dimensional SupervisedLearning

Mohammad Hossein Rohban, Prakash Ishwar , BirantOrten, William Karl , Venkatesh Saligrama

Achievable Rates for Intermittent Multi-AccessCommunication

Mostafa Khoshnevisan, J. Nicholas Laneman

17:10 Nonbinary LDPC-Coded Differential Modulation:Performance and Decoding Algorithm

Minghua Li , Bao-Ming Bai , Zhang Pei ying, Xiao Ma

Information-theoretic Limits on the Classification ofGaussian Mixtures: Classification on the GrassmannManifold

Matthew Nokleby , Robert Calderbank , Miguel Rodrigues

Gaussian Many-Access Channels: Definition andSymmetric Capacity

Dongning Guo, Xu Chen

17:30 Design of Non-Binary Quasi-Cyclic LDPC Codes by ACEOptimization

Alex Bazarsky , Noam Presman, Simon Litsyn

Asymptotically Minimax Regret by Bayes Mixtures forNon-exponential Families

Junichi Takeuchi , Andrew R. Barron

MAC Capacity Under Distributed Scheduling of MultipleUsers and Linear Decorrelation

Joseph Kampeas, Asaf Cohen, Omer Gurewitz

17:50 Improved decoding for binary source coding with codedside information

Anne Savard , Claudio Weidmann

Asymptotically equivalent sequences of matrices andrelative entropy

Jesus Gutierrez-Gutierrez, Pedro M. Crespo

On the Weakest Resource for Coordination in AV-MACswith Conferencing Encoders

Moritz Wiese, Holger Boche

20:45–22:00

Guided Tour of Seville’s Alcazar Palace∗

Meeting Point: Real Alcazar (Puerta del Leon)

∗ Tuesday’s visit to the Alcazar Palace does not take place in the conference venue, but in the city centre.

14


Wednesday, September 11

08:40–09:40

Integrating Diverse Types of Data to Search for New Therapeutic Strategies for DiseasesErnest Fraenkel (Massachusetts Institute of Technology, USA) (Location: Salon de Actos)

09:50–11:10

WeB1Real Number Codes

WeB2Inference, Information and Complexity

WeB3Network Coding


Chair: Jean-Claude Belfiore (Ecole Nationale Superieuredes Tlcommunications, France)

Chair: Ioannis Kontoyiannis (Athens UniversityEcon &Business, Greece)

Chair: Terence H. Chan (University of South Australia,Australia)

09:50 A new design criterion for spherically-shaped divisionalgebra-based space-time codes

Laura Luzzi , Roope Vehkalahti

Reconstruction in the Labeled Stochastic Block Model

Marc Lelarge, Laurent Massoulie, Jiaming Xu

Linear Network Coding for Multiple Groupcast Sessions:An Interference Alignment Approach

Abhik Kumar Das, Siddhartha Banerjee, SriramVishwanath

10:10 Projections, Dissections and Bandwidth ExpansionMappings

Antonio Campello, Vinay A. Vaishampayan, Sueli I. R.Costa

Information-Preserving Markov Aggregation

Bernhard C. Geiger , Christoph Temmel

On the capacity of ms/3t and 3s/nt sum-networks

Brijesh Kumar Rai , Niladri Das

10:30 Robust error correction for real-valued signals viamessage-passing decoding and spatial coupling

Jean Barbier , Florent Krzakala, Lenka Zdeborova, PanZhang

On the Noise Sensitivity and Mutual Information of(Nested-) Canalizing Boolean Functions

Johannes Georg Klotz, Martin Bossert , Steffen Schober

Combinatorial Flow over Cyclic Linear Networks

Chung Chan, Kenneth W. Shum, Qifu T Sun

10:50 Packing Tubes on Tori: An Efficient Method for Low SNRAnalog Error Correction

Robert M. Taylor, Jr., Lamine Mili , Amir Zaghloul

Coupled Neural Associative Memories

Amin Karbasi , Amir Hesam Salavati , Amin Shokrollahi

Outer Bounds and a Functional Study of the EdgeRemoval Problem

Eun Jee Lee, Michael Langberg, Michelle Effros

11:30–12:50

WeC1Lattice Codes

WeC2Entropy and Statistics of Sequences

WeC3Cooperation and Interference


Chair: Ramji Venkataramanan (University of Cambridge,United Kingdom)

Chair: Neri Merhav (Technion, Israel) Chair: Pablo Piantanida (Supelec, France)

11:30 Lattice Codes Based on Product Constructions over F2qwith Applications to Compute-and-Forward

Yu-Chih Huang, Krishna Narayanan

Information Theory for Atypical Sequences

Anders Høst-Madsen, Elyas Sabeti , Chad Walton

Cooperative Energy Harvesting Communications withRelaying and Energy Sharing

Kaya Tutuncuoglu, Aylin Yener

11:50 Precoded Integer-Forcing Universally Achieves the MIMOCapacity to Within a Constant Gap

Or Ordentlich, Uri Erez

A Phase Transition for the Uniform Distribution in thePattern Maximum Likelihood Problem

Winston Fernandes, Navin Kashyap

Zero vs. ε Error in Interference Channels

Ilia Levi , Danny Vilenchik , Michael Langberg, MichelleEffros

12:10 Enabling Multiplication in Lattice Codes via Construction A

Frederique Oggier , Jean-Claude Belfiore

Shannon Entropy Estimation from Convergence Results inthe Countable Alphabet Case

Jorge F. Silva, Patricio Parada

The Sum-Capacity of different K-user CognitiveInterference Channels in Strong Interference

Diana Maamari , Natasha Devroye, Daniela Tuninetti

12:30 Spatially-Coupled Low Density Lattices based onConstruction A with Applications to Compute-and-Forward

Nihat Tunali , Krishna Narayanan, Henry D. Pfister

The Entropy of Sums and Rusza’s Divergence on AbelianGroups

Ioannis Kontoyiannis, Mokshay Madiman

On Degrees-of-Freedom of Full-Duplex Uplink/DownlinkChannel

Achaleshwar Sahai , Suhas Diggavi , Ashutosh Sabharwal

14:00–14:30

Information and Control in Sensing-Acting Systems: Predictive Information and the Emergence of Hierarchical RepresentationsNaftali Tishby (The Hebrew University, Israel) (Location: Salon de Actos)

14:30–15:00

Source Coding, Lists, and Renyi EntropyAmos Lapidoth (ETHZ, Switzerland) (Location: Salon de Actos)

15:00–15:30

The Role of the Hypercontractivity Ribbon in Information TheoryVenkat Anantharam (University of California, Berkeley, USA) (Location: Salon de Actos)

15


Thursday, September 12

08:40–09:40

Growth Optimal Empirical Portfolio Selection StrategiesLaszlo Gyorfi (Budapest University of Technology and Economics, Hungary) (Location: Salon de Actos)

09:50–11:10

ThB1Algebraic Codes

ThB2Statistics and Learning(Organized by Gabor Lugosi)

ThB3Interference and Relay Channels


Chair: Navin Kashyap (Indian Institute of Science, India) Chair: Gabor Lugosi (Universitat Pompeu Fabra, Spain) Chair: Ayfer Ozgur (Stanford University, USA)

09:50 Computing the Camion’s multivariate BCH bound

Jose Joaquın Bernal , Diana Bueno-Carreno, Juan Simon

From Bandits to Experts: A Tale of Domination andIndependence

Noga Alon, Nicolo Cesa-Bianchi , Claudio Gentile, YishayMansour

Multilevel Topological Interference Management

Chunhua Geng, Hua Sun, Syed Ali Jafar

10:10 Flag Orbit Codes and Their Expansion to Stiefel Codes

Renaud-Alexandre Pitaval , Olav Tirkkonen

On Semi-Probabilistic Universal Prediction

Alexander Rakhlin, Karthik Sridharan

On the Interference Channel with Common Messages andthe Role of Rate-Sharing

Stefano Rini , Andrea Goldsmith

10:30 New Geometrical Spectra of Linear Codes withApplications to Performance Analysis

Xiao Ma, Jia Liu, Qiutao Zhuang, Bao-Ming Bai

Learning Joint Quantizers for Reconstruction andPrediction

Maxim Raginsky

The Stability Region of the Two-User Interference Channel

Nikolaos Pappas, Marios Kountouris, Anthony Ephremides

10:50 Concatenated Permutation Block Codes based on SetPartitioning for Substitution and Deletion Error-Control

Reolyn Heymann, Jos H. Weber , Theo G. Swart , HendrikC. Ferreira

Prediction of Individual Sequences with Finite-MemoryUniversal Predictors

Meir Feder

On the Reliable Transmission of Correlated Sources OverTwo-Relay Network

Mohammad Nasiraee, Bahareh Akhbari , MahmoudAhmadian, Mohammad Reza Aref

11:30–12:50

ThC1Source Coding

ThC2Information and Estimation

ThC3Interference Networks


Chair: Ram Zamir (Tel Aviv University, Israel) Chair: Laszlo Gyorfi (Budapest University of Technologyand Economics, Hungary)

Chair: Syed Ali Jafar (University of California Irvine, USA)

11:30 Nonasymptotic Noisy Source Coding

Victoria Kostina, Sergio Verdu

Non-Parametric Prediction of the Mid-Price Dynamics in aLimit Order Book

Deepan Palguna, Ilya Pollak

Degrees of Freedom of the Rank-Deficient MIMO Xchannel

Adrian Agustin, Josep Vidal

11:50 The Heegard-Berger Problem with Common ReceiverReconstructions

Badri N. Vellambi , Roy Timo

One-shot bounds for various information theoreticproblems using smooth min and max Renyi divergences

Naqueeb Warsi

Interference Neutralization using Lattice Codes

Shahab Ghasemi-Goojani , Hamid Behroozi

12:10 Lattice Quantization Noise Revisited

Cong Ling, Lu Gan

Signals that can be easily time-frequency synchronizedfrom their ambiguity function

Tohru Kohda, Yutaka Jitsumatsu, Kazuyuki Aihara

Cognitive Cooperative Communications on the MultipleAccess Channel

Jonathan Shimonovich, Anelia Somekh-Baruch, Shlomo(Shitz) Shamai

12:30 The Likelihood Encoder for Source Coding

Paul Cuff , Eva Chen Song

Novel Tight Classification Error Bounds under MismatchConditions based on f -Divergence

Ralf Schluter , Markus Nussbaum-Thom, Eugen Beck ,Tamer Alkhouli , Hermann Ney

State-Dependent Gaussian Z-Channel with MismatchedSide-Information and Interference

Ruchen Duan, Yingbin Liang, Ashish Khisti , Shlomo(Shitz) Shamai

16


Thursday, September 12 (Cont’d)

14:00–14:30

Detection of Correlations and Random Geometric GraphsGabor Lugosi (Universitat Pompeu Fabra, Spain) (Location: Salon de Actos)

14:30–15:00

Coding for Chip-to-Chip CommunicationAmin Shokrollahi (Kandou Bus and EPFL, Switzerland) (Location: Salon de Actos)

15:10–16:30

ThE1Codes for Storage

ThE2Big Data: Statistics and InformationTheory in High Dimensions andUndersampled Regimes(Organized by Prasad Santhanam)

ThE3Source Coding in Networks


Chair: Hendrik C. Ferreira (University of Johannesburg,South Africa)

Chair: Prasad Santhanam (University of Hawaii at Manoa,USA)

Chair: J. Nicholas Laneman (University of Notre Dame,USA)

15:10 Exact-Regenerating Codes between MBR and MSR Points

Toni Ernvall

Informational Confidence Bounds for Self-NormalizedAverages and Applications

Aurelien Garivier

On Existence of Optimal Linear Encoders over Non-fieldRings for Data Compression with Application to Computing

Sheng Huang, Mikael Skoglund

15:30 Prefixless q-ary Balanced Codes with ECC

Theo G. Swart , Kees A. Schouhamer Immink

MCUIUC — A New Framework for Metagenomic ReadCompression

Jonathan Ligo, Minji Kim, Amin Emad , Olgica Milenkovic,Venugopal Veeravalli

Sampling versus Random Binning for Multiple Descriptionsof a Bandlimited Source

Adam Mashiach, Jan Østergaard , Ram Zamir

15:50 Constrained Rank Modulation Schemes

Frederic Sala, Lara Dolecek

Sparse Regression Codes: Recent Results and FutureDirections

Ramji Venkataramanan, Sekhar Tatikonda

Capacity Region of Multi-Resolution Streaming inPeer-to-Peer Networks

Batuhan Karagoz, Semih Yavuz, Tracey Ho, MichelleEffros

16:10 Maximum Likelihood Associative Memories

Vincent Gripon, Michael Rabbat

Estimation of transition and stationary probabilities of slowmixing Markov processes

Narayana Prasad Santhanam, Meysam Asadi , RamezanParavi Torghabeh

A Deterministic Annealing Approach to Optimization ofZero-delay Source-Channel Codes

Mustafa S. Mehmetoglu, Emrah Akyol , Kenneth Rose

16:50–18:10

ThF1Source and Channel Coding

ThF2Information Theoretic Problems inComputer Science(Organized by Venkat Anantharam)

ThF3Multicell and Broadcast


Chair: Ashish Khisti (University of Toronto, Canada) Chair: Venkat Anantharam (University of California atBerkeley, USA)

Chair: Giuseppe Caire (University of Southern California,USA)

16:50 The Least Degraded and the Least Upgraded Channel withrespect to a Channel Family

Wei Liu, S. Hamed Hassani , Rudiger Urbanke

Bypassing Correlation Decay for Matchings with anApplication to XORSAT

Marc Lelarge

Uplink Multi-Cell Processing: Approximate Sum Capacityunder a Sum Backhaul Constrain

Yuhan Zhou, Wei Yu, Dimitris Toumpakaris

17:10 On 2-D Non-Adjacent-Error Channel Models

Shivkumar Manickam, Navin Kashyap

Lovasz ϑ, SVMs and Applications

Vinay Jethava, Jacob Sznajdman, Chiranjib Bhattacharya,Devdatt Dubhashi

Multi-Cell Cooperation with Random User Locations underArbitrary Signaling

Maksym A. Girnyk , Mikko Vehkapera, Lars K. Rasmussen

17:30 On the Mutual Information between Random Variables inNetworks

Xiaoli Xu, Satyajit Thakor , Yong Liang Guan

Local Privacy, Quantitative Data Processing Inequalities,and Statistical Minimax Rates

John Duchi , Michael Jordan, Martin J. Wainwright

Any Positive Feedback Rate Increases the Capacity ofStrictly Less-Noisy Broadcast Channels

Youlong Wu, Michele Wigger

17:50 A Connection between Good Rate-distortion Codes andBackward DMCs

Curt Schieler , Paul Cuff

On the Information Complexity of Cascaded Norms withSmall Domains

T. S. Jayram

On the Role of Interference Decoding in CompoundBroadcast Channels

Meryem Benammar , Pablo Piantanida

20:30–23:00

Conference Banquet∗Location: Restaurante Abades (C/ Betis, 69)

∗ The conference banquet does not take place in the conference venue, but in the Triana district, in the city centre.

17


Friday, September 13

08:40–09:40

Strategic Computation in NetworksMichael Kearns (University of Pennsylvania, USA) (Location: Salon de Actos)

09:50–11:10

FrB1Information Function Computation

FrB2Simulation and Approximation

FrB3Relay Networks


Chair: Maxim Raginsky (University of Illinois atUrbana-Champaign, USA)

Chair: Paul Cuff (Princeton University, USA) Chair: Michele A. Wigger (Telecom ParisTech, France)

09:50 Distributed Function Computation Over a Tree Network

Milad Sefidgaran, Aslan Tchamkerten

When is it possible to simulate a DMC channel fromanother?

Farzin Haddadpour , Mohammad Hossein Yassaee,Mohammad Reza Aref , Amin Aminzadeh Gohari

On the Stability Region of a Relay-Assisted MultipleAccess Scheme

Nikolaos Pappas, Marios Kountouris, AnthonyEphremides, Apostolos Traganitis

10:10 Two-Partition-Symmetrical Entropy Function Regions

Qi Chen, Raymond W. Yeung

Joint Channel Intrinsic Randomness and ChannelResolvability

Alexandre J. Pierrot , Matthieu R. Matthieu R.

On information flow and feedback in relay networks

Bobbie Chern, Ayfer Ozgur

10:30 Bounding the Entropic Region via Information Geometry

Yunshu Liu, John Maclaren Walsh

Fixed-to-Variable Length Resolution Coding for TargetDistributions

Georg Bocherer , Rana Ali Amjad

Two-Unicast Two-Hop Interference Network: Finite-FieldModel

Song-Nam Hong, Giuseppe Caire

10:50 Characterization of the Smooth Renyi Entropy UsingMajorization

Hiroki Koga

A New Unified Method for Intrinsic Randomness Problemsof General Sources

Tomohiko Uyematsu, Shohei Kunimatsu

Cyclic Interference Neutralization on the 2 × 2 × 2Full-Duplex Two-Way Relay-Interference Channel

Henning Maier , Rudolf Mathar

11:30–12:50

FrC1Error Exponent and Capacity

FrC2Topics in Compression

FrC3Capacity of Relay Networks


Chair: Tobias Koch (Universidad Carlos III de Madrid,Spain)

Chair: Claudio Weidmann (CNRS / ENSEA / UniversityCergy-Pontoise, France)

Chair: Nikolaos Pappas (Supelec, France)

11:30 On the Average-Listsize Capacity and the Cutoff Rate ofDiscrete Memoryless Channels with Feedback

Christoph Bunte, Amos Lapidoth

Compression of Noisy Signals with Information Bottlenecks

Amin Emad , Olgica Milenkovic

Improved Capacity Approximations for Gaussian RelayNetworks

Ritesh Kolte, Ayfer Ozgur

11:50 Improved Capacity Lower Bounds for Channels withDeletions and Insertions

Ramji Venkataramanan, Sekhar Tatikonda

Distortion Minimization in Layered Broadcast Transmissionof a Gaussian Source Over Rayleigh Channels

Wessam Mesbah, Mohammad Shaqfeh, Hussein Alnuweiri

Gaussian Half-Duplex Relay Networks: Improved Gap anda Connection with the Assignment Problem

Martina Cardone, Daniela Tuninetti , Raymond Knopp,Umer Salim

12:10 ε-Capacity and Strong Converse for Channels withGeneral State

Marco Tomamichel , Vincent Y. F. Tan

Sparse Signal Processing with Linear and Non-LinearObservations: A Unified Shannon Theoretic Approach

Cem Aksoylar , George Atia, Venkatesh Saligrama

Capacity Region of a Class of Interfering Relay Channels

Hieu T. Do, Tobias J. Oechtering, Mikael Skoglund , Mai Vu

12:30 Rate-Distortion Bounds for an Epsilon-InsensitiveDistortion Measure

Kazuho Watanabe

The Deterministic Capacity of Relay Networks with RelayPrivate Messages

Ahmed A. Zewail , Yahya Mohasseb, Mohammed Nafie,Hesham El Gamal

18

Abstracts

Pantone© 137

Tintas planas

Cuatricromía

Pantone© 2738

C=0 M=50 Y=100 K=0

C=100 M=100 Y=0 K=0


Monday

Claude Shannon (by Sergio Verdu)

Monday, 18:45–19:45 Paraninfo Univ. de Sevilla (C/ San Fernando 4)

Chair: Fernando Perez-Cruz (Universidad Carlos III de Madrid, Spain)

Welcome Reception

Monday, 20:30–22:15 Hotel Los Seises (C/ Segovias, 6)

Tuesday

Tuesday’s Plenary, 08:40–09:40

Tuesday, 08:40–09:40 Salon de Actos

Chair: Albert Guillen i Fabregas (ICREA and Universitat Pompeu Fabra,Spain)

From Statistics to Information Theory: Challenges in the BigData EraMartin J. Wainwright (University of California, Berkeley, USA)

While statistics and information theory have a lengthy shared history, the modernera of massive data sets leads to various new challenges. In this talk, we discussseveral problem domains which have witnessed fruitful interactions between thetwo fields, including large-scale social network analysis, nonparametric regressionand sparsity in high dimensions, and oracle complexity of convex optimization.Such areas provide opportunities for information-theoretic tools to be brought tobear in interesting and non-orthodox ways.

Bio: Martin Wainwright is currently a professor at University of California atBerkeley, with a joint appointment between the Department of Statistics and theDepartment of Electrical Engineering and Computer Sciences (EECS). He receiveda Bachelor’s degree in Mathematics from University of Waterloo, Canada, andPh. D. degree in EECS from Massachusetts Institute of Technology (MIT). His re-search interests include high-dimensional statistics, statistical machine learning, in-formation theory and statistical signal processing. He has been awarded the GeorgeM. Sprowls Prize for his dissertation research (MIT), an Alfred P. Sloan Foun-dation Fellowship, Best Paper Awards from the IEEE Signal Processing Society(2008), IEEE Communications Society (2010), the Joint Paper Prize (2012) fromIEEE Information Theory and Communication Societies, and a Medallion Lecturer(2013) from the Institute of Mathematical Statistics.

Tuesday, 09:50–11:10

Tuesday, 09:50–11:10 Room 1 – Sala Juan Larraneta

TuB1: Polar Coding 1

Chair: Joerg Kliewer (New Mexico State University, USA)

09:50

Polar Codes with Dynamic Frozen Symbols and Their Decod-ing by Directed SearchPeter Trifonov (Saint-Petersburg State Polytechnic University, Russia);

Vera Miloslavskaya (Saint-Petersburg State Polytechnic University, Rus-sia)

A novel construction of polar codes with dynamic frozen symbols is proposed.The proposed codes are subcodes of extended BCH codes, which ensure suffi-ciently high minimum distance. Furthermore, a decoding algorithm is proposed,which employs estimates of the not-yet-processed bit channel error probabilities toperform directed search in code tree, reducing thus the total number of iterations.

10:10

A Lower Bound on Achievable Rates by Polar Codes with Mis-match Polar DecodingMine Alsan (EPFL, Switzerland)

In this paper we show that mismatched polar codes over B-DMCs can achievepositive rates of at least I(W,V ) whenever I(W,V ) > 0, where W denotes thecommunication channel, V the mismatched channel used in the code design, and

I(W,V ) =∑

y

∑

x∈0,1

1

2W (y|x) log

V (y|x)12V (y|0) + 1

2V (y|1)

.

10:30

Scaling Exponent of List Decoders with Applications to PolarCodesMarco Mondelli (EPFL, Switzerland); S. Hamed Hassani (EPFL,Switzerland); Rudiger Urbanke (EPFL, Switzerland)

Motivated by the significant performance gains which polar codes experience whenthey are decoded with successive cancellation list decoders, we study how the scal-ing exponent changes as a function of the list size L. In particular, we fix the blockerror probability Pe and we analyze the tradeoff between the blocklength N andthe back-off from capacity C − R using scaling laws. By means of a Divide andIntersect procedure, we provide a lower bound on the error probability under MAPdecoding with list size L for any binary-input memoryless output-symmetric chan-nel and for any class of linear codes such that their minimum distance is unboundedas the blocklength grows large. We show that, although list decoding can signifi-cantly improve the involved constants, the scaling exponent itself, i.e., the speed atwhich capacity is approached, stays unaffected. This result applies in particular topolar codes, since their minimum distance tends to infinity as N increases. Someconsiderations are also pointed out for the genie-aided successive cancellation de-coder when transmission takes place over the binary erasure channel.

10:50

Polar Coding for Fading ChannelsHongbo Si (The University of Texas at Austin, USA); Onur Ozan Koylu-oglu (The University of Texas at Austin, USA); Sriram Vishwanath (Uni-versity of Texas at Austin, USA)

A polar coding scheme for fading channels is proposed in this paper. More specifi-cally, the focus is on the Gaussian fading channel with a BPSK modulation, wherethe equivalent channel is modeled as a binary symmetric channel with varyingcross-over probabilities. To deal with variable channel states, a coding schemeof hierarchically utilizing polar codes is proposed. In particular, by observing thepolarization of different binary symmetric channels over different fading blocks,each channel use corresponding to a different polarization is modeled as a binaryerasure channel such that polar codes could be adopted to encode over blocks. It isshown that the proposed coding scheme, without instantaneous channel state infor-mation at the transmitter, achieves the capacity of the corresponding fading binarysymmetric channel.

Tuesday, 09:50–11:10 Room 2 – Sala de Grados

TuB2: Networks: Compression and Coding(Organized by Suhas Diggavi and Michelle Effros)

Chair: Michelle Effros (California Institute of Technology, USA)

09:50

On Symmetric Multiple Description CodingLin Song (McMaster University, Canada); Shuo Shao (McMaster Univer-

21


sity, Canada); Jun Chen (McMaster University, Canada)

We derive a single-letter lower bound on the minimum sum rate of multiple descrip-tion coding with symmetric distortion constraints. For the binary uniform sourcewith the Hamming distortion measure, this lower bound can be evaluated with theaid of a certain minimax theorem. A similar minimax theorem is established inthe quadratic Gaussian setting, which is further leveraged to analyze the specialcase where the minimum sum rate subject to two levels of distortion constraints(with the second level imposed on the complete set of descriptions) is attained; inparticular, we determine the minimum achievable distortions at the intermediatelevels.

10:10

Characterising Correlation via Entropy FunctionsSatyajit Thakor (Institute of Network Coding, The Chinese University ofHong Kong, Hong Kong); Terence H. Chan (University of South Aus-tralia, Australia); Alex Grant (University of South Australia, Australia)

Characterising the capacity region for a network can be extremely difficult. Evenwith independent sources, deter- mining the capacity region can be as hard as theopen problem of characterising all information inequalities. The majority of com-putable outer bounds in the literature are relaxations of the Linear Programmingbound which involves entropy functions of random variables related to the sourcesand link messages. When sources are not independent, the problem is even morecompli- cated. Extension of Linear Programming bounds to networks with cor-related sources is largely open. Source dependence is usually specified via a jointprobability distribution, and one of the main challenges in extending linear programbounds is the difficulty (or impossibility) of characterising arbitrary depen- denciesvia entropy functions. This paper tackles the problem by answering the question ofhow well entropy functions can characterise correlation among sources. We showthat by using carefully chosen auxiliary random variables, the characterisation canbe fairly “accurate”.

10:30

Erasure/List Exponents for Slepian-Wolf DecodingNeri Merhav (Technion, Israel)

We analyze random coding error exponents associated with erasure/list Slepian-Wolf decoding using two different methods and then compare the resulting bounds.The first method follows the well known techniques of Gallager and Forney and thesecond method is based on a technique of distance enumeration, or more generally,type class enumeration, which is rooted in the statistical mechanics of a disorderedsystem that is related to the random energy model (REM). The second method isguaranteed to yield exponent functions which are at least as tight as those of thefirst method, and it is demonstrated that for certain combinations of coding ratesand thresholds, the bounds of the second method are strictly tighter than those ofthe first method, by an arbitrarily large factor. In fact, the second method may evenyield an infinite exponent at regions where the first method gives finite values. Wealso discuss the option of variable-rate Slepian-Wolf encoding and demonstratehow it can improve on the resulting exponents.

10:50

Operational Extremality of Gaussianity in Network Compres-sion, Communication, and CodingHimanshu Asnani (Stanford University, USA); Ilan Shomorony (CornellUniversity, USA); Salman Avestimehr (Cornell University, USA); TsachyWeissman (Stanford University, USA)

Among other extremal properties, Gaussian sources are hardest to compress andcommunicate over. We review the main results of Shomorony and Avestimehr, andAsnani, Shomorony, Avestimehr, and Weissman, exhibiting the generality in whichsuch extremal properties hold in compression, communication and coding overnetworks. These properties are established via operational arguments, bypassingelusive characterizations of fundamental performance limits: schemes tailored forthe Gaussian case are harnessed for constructions of schemes that provably doessentially as well under any other source of the same covariance. The talk willhighlight the main ideas behind these constructions and how the results, whichwere established for memoryless sources and channels, carry over to the presenceof memory.

Tuesday, 09:50–11:10 Room 3 – Aula 304

TuB3: Secrecy, Privacy and Cryptography

Chair: Matthieu Bloch (Georgia Institute of Technology, France)

09:50

Spectrum Bandit OptimizationMarc Lelarge (INRIA and ENS, France); Alexandre Proutiere (MicrosoftResearch, United Kingdom); M. Sadegh Talebi (KTH Royal Institute ofTechnology, Sweden)

We consider the problem of allocating radio channels to links in a wireless net-work. Links interact through interference, modelled as a conflict graph (i.e., twointerfering links cannot be simultaneously active on the same channel). We aim atidentifying the channel allocation maximizing the total network throughput over afinite time horizon. Should we know the average radio conditions on each channeland on each link, an optimal allocation would be obtained by solving an IntegerLinear Program (ILP). When radio conditions are unknown a priori, we look for asequential channel allocation policy that converges to the optimal allocation whileminimizing on the way the throughput loss or regret due to the need for explor-ing suboptimal allocations. We formulate this problem as a generic linear banditproblem, and analyze it in a stochastic setting where radio conditions are drivenby a i.i.d. stochastic process, and in an adversarial setting where radio conditionscan evolve arbitrarily. We provide, in both settings, algorithms whose regret upperbounds outperform those of existing algorithms.

10:10

Physical-Layer Cryptography Through Massive MIMOThomas Dean (Stanford University, USA); Andrea Goldsmith (StanfordUniversity, USA)

We propose the new technique of physical-layer cryptography based on using amassive MIMO channel as a key between the sender and desired receiver, whichneed not be secret. The goal is for low-complexity encoding and decoding by thedesired transmitter-receiver pair, whereas decoding by an eavesdropper is hard interms of prohibitive complexity. The massive MIMO system has a channel gainmatrix that is drawn i.i.d. according to a Gaussian distribution, subject to additivewhite Gaussian noise. The decoding complexity is analyzed by mapping the mas-sive MIMO system to a lattice. We show that the eavesdropper’s decoder for theMIMO system with M-PAM modulation is equivalent to solving standard latticeproblems that are conjectured to be of exponential complexity for both classicaland quantum computers. Hence, under the widely-held conjecture that standardlattice problems are of worst-case complexity, the proposed encryption scheme hassecurity that exceeds that of the most common encryption methods used today suchas RSA and Diffie-Hellman. Additionally, we show that this scheme could be usedto securely communicate without a pre-shared secret key and little computationaloverhead. In particular, a standard parallel channel decomposition allows the de-sired transmitter-receiver pair to encode and decode transmissions over the MIMOchannel based on the singular value decomposition of the channel, while decodingremains computationally hard for an eavesdropper with an independent channelgain matrix, even if it knows the channel gain matrix between the desired trans-mitter and receiver. Thus, the massive MIMO system provides for low-complexityencryption commensurate with the most sophisticated forms of application-layerencryption by exploiting the physical layer properties of the radio channel.

10:30

Secrecy & Rate Adaptation for Secure HARQ ProtocolsMael Le Treust (INRS, Canada); Leszek Szczecinski (INRS-EMT,Canada); Fabrice Labeau (McGill University, Canada)

This paper is dedicated to the study of HARQ protocols under a secrecy constraint.An encoder sends information to a legitimate decoder while keeping it secret fromthe eavesdropper. Our objective is to provide a coding scheme that satisfies bothreliability and confidentiality conditions. This problem has been investigated inthe literature using a coding scheme that involves a unique secrecy parameter. Theuniqueness of this parameter is sub-optimal for the throughput criteria and we pro-pose a new coding scheme that introduces additional degrees of freedom. Our codeinvolves Secrecy Adaptation and Rate Adaptation and we called it SARA-code.The first contribution is to prove that the SARA-code has small error probabilityand small information leakage rate. The second contribution is to show, over a

22


numerical example, that the SARA-code improves the secrecy throughput.

10:50

Protecting Data Against Unwanted InferencesSupriyo Chakraborty (University of California at Los Angeles, USA);Nicolas Bitouze (University of California, Los Angeles, USA); ManiB. Srivastava (University of California, Los Angeles, USA); Lara Dolecek(UCLA, USA)

We study the competing goals of utility and privacy as they arise when a providerdelegates the processing of its personal information to a recipient who is better ableto handle this data. We formulate our goals in terms of the inferences which canbe drawn using the shared data. A whitelist describes the inferences that are desir-able, i.e., providing utility. A blacklist describes the unwanted inferences which theprovider wants to keep private. We formally define utility and privacy parametersusing elementary information-theoretic notions and derive a bound on the regionspanned by these parameters. We provide constructive schemes for achieving cer-tain boundary points of this region. Finally, we improve the region by sharing dataover aggregated time slots.

Tuesday, 11:30–12:50


TuC1: Polar Coding 2

Chair: Olgica Milenkovic (University of Illinois, USA)

11:30

Polar Coding for Secret-Key GenerationRemi A. Chou (Georgia Institute of Technology, USA); Matthieu R. Bloch(Georgia Institute of Technology, France); Emmanuel Abbe (PrincetonUniversity, USA)

Practical implementations of secret-key generation are often based on sequentialstrategies, which handle reliability and secrecy in two successive steps, called rec-onciliation and privacy amplification. In this paper, we propose an alternativescheme based on polar coding that jointly deals with reliability and secrecy. Westudy a binary degraded symmetric discrete memoryless source model with uni-form marginals, and assume one-way rate-limited public communication betweentwo legitimate users. Specifically, we propose secret-key capacity-achieving polarcoding schemes, in which users rely on pre-shared secret seed of negligible rate.For the model studied, we thus provide the first example of low-complexity secret-key capacity-achieving scheme that handles vector quantization, for rate-limitedpublic communication. Furthermore, we provide examples for which no seed isrequired.

11:50

On Polarization for the Linear Operator ChannelCesar Brito (New Mexico State University, USA); Joerg Kliewer (NewMexico State University, USA)

We address the problem of reliably transmitting information through a networkwhere the nodes perform random linear network coding and where an adversarypotentially injects malicious packets into the network. A good model for such achannel is a linear operator channel, where in this work we employ a combinedmultiplicative and additive matrix channel. We show that this adversarial chan-nel behaves like a subspace-based symmetric discrete memoryless channel (DMC)under subspace insertions and deletions and typically has an input alphabet withnon-prime cardinality. This facilitates the recent application of channel polariza-tion results for DMCs with arbitrary input alphabets by providing a suitable one-to-one mapping from input matrices to subspaces. As a consequence, we showthat polarization for this adversarial linear operator channel can be obtained viaan element-wise encoder mapping for the input matrices, which replaces the finitefield summation in the channel combining step for Arıkan’s classical polar codes.

12:10

Channel Polarization with Higher Order Memory

Huseyin Afser (University of Bogazici, Turkey); Hakan Delic (BogaziciUniversity, Turkey)

We introduce the design of a class of code sequences C(m)n , n = 1, 2, . . . with

memory levelm = 1, 2, . . . based on the channel polarization idea, where C(1)n coincides with the polar codes presented by Arıkan in [1]. The new codes achievethe symmetric capacity of arbitrary binary-input discrete memoryless channels. Wederive bounds on the polarization performance as scaled with m. We show thatC(m)n offers monotonically decreasing encoding and decoding complexity with

growing m.


TuC2: IT & Coding for Contemporary Video(Organized by Christina Fragouli and Emina Soljanin)

Chair: Emina Soljanin (Bell Labs, Alcatel - Lucent, USA)

11:30

Some Coding and Information Theoretic Problems in Contem-porary (Video) Content DeliveryEmina Soljanin (Bell Labs, Alcatel - Lucent, USA)

Information and coding theory have traditionally been used in point-to-point sce-narios, to compute and achieve the transmission channel capacity as well as tocompute and achieve the optimal compression rate vs. source distortion trade-off.In today’s networks, the same (video) data is often transmitted to multiple users,simultaneously over diverse channels. The users may differ not only in the sizeand resolution of their displays and computing power, but may also be interestedin different video scenes, with different levels of distortion, or even with differentdistortion measures. This paper describes several transmission and compressionproblems that arise in such heterogeneous network scenarios, and discusses howinformation and coding theory could be used to address them.

11:50

Source Broadcasting over Erasure Channels: DistortionBounds and Code DesignLouis Tan (University of Toronto, Canada); Yao Li (UCLA, USA); AshishKhisti (University of Toronto, Canada); Emina Soljanin (Bell Labs, Alca-tel - Lucent, USA)

We study a lossy source-broadcasting problem involving the transmission of a bi-nary source over a two-receiver erasure broadcast channel. The motivation of ourwork stems from the problem faced by a server that wishes to broadcast contentto a diverse set of users with fractional source reconstruction requirements. In thisproblem, the server wishes to minimize the overall network latency incurred (mea-sured by the number of channel uses per source symbol) when faced with users ofheterogeneous channel qualities, computing capabilities, content demand etc. Weprovide two complementary approaches to this problem. The first approach is toconsider the problem from a joint source-channel coding formulation. Under thisformulation, we provide both inner and outer bounds for the network latency underan erasure distortion criterion. Alternatively, the second approach employs ratelesscoding and formulates an optimization problem so as to find a degree distributionthat minimizes the network latency. We compare both approaches with numericalsimulations.

12:10

Impact of Random and Burst Packet Losses on H.264 ScalableVideo CodingSiyu Tang (Alcatel-Lucent Bell Labs, Belgium); Patrice Rondao Alface(Alcatel-Lucent Bell Labs, Belgium)

This paper studies the impact of packet loss on H.264 scalable video coding (SVC).A Markov Chain (MC) with 2N states is developed to describe the error propaga-tion process inside a group of pictures (GOP). The characteristic of different packetloss events is captured by the initial state vector of the MC. Based on the proposedmodel, the performance of different GOP structures can be evaluated under bothrandom and burst packet losses.

12:30

23


Network Coding Designs Suited for the Real World: Whatworks, What doesn’t, What’s PromisingMorten V. Pedersen (Aalborg University, Denmark); Daniel E. Lucani(Aalborg University, Denmark); Frank H.P. Fitzek (Aalborg University,Denmark); Chres Sørensen (Aalborg University, Denmark); Arash Shah-baz Badr (Technische Universitat Hamburg–Harburg, Germany)

Network coding (NC) has attracted tremendous attention from the research com-munity due to its potential to significantly improve networks’ throughput, delay,and energy performance as well as a means to simplify protocol design and natu-rally providing security support. The possibilities in code design have produced alarge influx of new ideas and approaches to harness the power of NC. But, which ofthese designs are truly successful in practice? and which designs will not live up totheir promised theoretical gains due to real–world constraints? Without attemptinga comprehensive view of all practical pitfalls, this paper seeks to identify key ingre-dients to a successful design, critical and common limitations to most intra–sessionNC systems as well as promising techniques and ideas to guide future models andresearch problems grounded on practical concerns.


TuC3: Information-Theoretic Secrecy

Chair: Aylin Yener (Pennsylvania State University, USA)

11:30

Exploiting Common Randomness: a Resource for Network Se-crecyLaszlo Czap (Ecole Polytechinque Federale de Lausanne, EPFL, Switzer-land); Vinod M. Prabhakaran (Tata Institute of Fundamental Research,India); Suhas Diggavi (University of California Los Angeles, USA);Christina Fragouli (EPFL, Switzerland)

We investigate the problem of secure communication in a simple network withthree communicating parties, two distributed sources who communicate over or-thogonal channels to one destination node. The cooperation between the sourcesis restricted to a rate limited common random source they both observe. The com-munication channels are erasure channels with strictly causal channel state infor-mation of the destination available publicly. A passive adversary is present in thesystem eavesdropping on any one of the channels. We design a linear scheme thatensures secrecy against the eavesdropper. By deriving an outer bound for the prob-lem we prove that the scheme is optimal in certain special cases.

11:50

Secrecy in Cascade NetworksPaul Cuff (Princeton University, USA)

We consider a cascade network where a sequence of nodes each send a message totheir downstream neighbor to enable coordination, the first node having access toan information signal. An adversary also receives all the communication as wellas additional side-information. The performance of the system is measured by apayoff function over all actions produced by the nodes as well as the adversary.The challenge is to maintain secrecy from the adversary in order thwart his attemptto reduce the payoff. We obtain inner and outer bounds on performance, and giveexamples where they are tight. From these bounds, we also derive the optimalequivocation that can be achieved in this setting, as a special case.

12:10

Secure Degrees of Freedom of MIMO X-Channels with Out-put Feedback and Delayed CSIAbdellatif Zaidi (Universite Paris-Est Marne La Vallee, France); Zo-haib Hassan Awan (Universite Catholique de Louvain, Belgium); Shlomo(Shitz) Shamai (The Technion, Israel); Luc Vandendorpe (University ofLouvain, Belgium)

We investigate the problem of secure transmission over a two-user multi-inputmulti-output (MIMO) X-channel with noiseless local feedback and delayed chan-nel state information (CSI) available at transmitters. The transmitters are equippedwith M antennas each, and the receivers are equipped with N antennas each. For

this model, we characterize the optimal sum secure degrees of freedom (SDoF)region. We show that, in presence of local feedback and delayed CSI, the sumSDoF region of the MIMO X-channel is same as the SDoF region of a two-userMIMO BC with 2M antennas at the transmitter and N antennas at each receiver.This result shows that, upon availability of feedback and delayed CSI, there is noperformance loss in sum SDoF due to the distributed nature of the transmitters.Next, we show that this result also holds if only global feedback is conveyed tothe transmitters. We also study the case in which only local feedback is providedto the transmitters, i.e., without CSI, and derive a lower bound on the sum SDoFfor this model. Furthermore, we specialize our results to the case in which thereare no security constraints. In particular, similar to the setting with security con-straints, we show that the optimal sum degrees of freedom (sum DoF) region of the(M,M,N,N)-MIMO X-channel is same of the DoF region of a two-user MIMOBC with 2M antennas at the transmitter and N antennas at each receiver. Weillustrate our results with some numerical examples.

12:30

The Secrecy Capacity of a Compound MIMO Gaussian Chan-nelRafael F. Schaefer (Technische Universitat Munchen, Germany); SergeyLoyka (University of Ottawa, Canada)

The compound MIMO Gaussian wiretap channel is studied, where the channel tothe legitimate receiver is known and the eavesdropper channel is not known to thetransmitter but is known to have a bounded spectral norm (channel gain). The com-pound secrecy capacity is established without the degradedness assumption andthe optimal signaling is identified: the compound capacity equals the worst-casechannel capacity thus establishing the saddle-point property, the optimal signalingis Gaussian and on the eigenvectors of the legitimate channel and the worst-caseeavesdropper is isotropic. The eigenmode power allocation somewhat resemblesthe standard water-filling but is not identical to it.

Tuesday’s Plenaritas, 14:00–15:00

Tuesday, 14:00–15:00 Salon de Actos

Chair: Chair: Michael Honig (Northwestern University, USA)

14:00

Polar versus Spatial Coupling: The Good, the Bad, and theUglyRudiger Urbanke (EPFL, Switzerland)

Once upon a time it was deemed to be difficult to construct coding schemes thatallow reliable transmission close to capacity at low complexity. But these days wehave several such schemes. How do they compare? I will discuss in particular polarcodes and spatially coupled codes. Which wins when we consider their scalingbehavior, their complexity, the achievable throughput, universality, or robustness?As we will see, much is known, but many questions are still open.(Joint work with Hamed Hassani)

Bio: Rudiger Urbanke received the Diplomingenieur degree from the Vienna In-stitute of Technology, Vienna, Austria, in 1990 and the M. S. and Ph. D. degreesin electrical engineering from Washington University, St. Louis, MO, in 1992 and1995 respectively. From 1995 to 1999, he held a position at the Mathematics ofCommunications Department at Bell Labs. Since November 1999, he has been afaculty member at the School of Computer and Communication Sciences of EPFL,Lausanne, Switzerland. Dr. Urbanke is a recipient of a Fulbright Scholarship. In2000–2004 he was an Associate Editor of the IEEE Transactions on InformationTheory and he is currently on the board of the series ‘Foundations and Trendsin Communications and Information Theory’. He is a co-recipient of the IEEEInformation Theory Society 2002 Best Paper Award and of the 2011 IEEE KojiKobayashi award. He co-authored the book Modern Coding Theory published byCambridge University Press.

14:30

Managing Bursty Interference with FeedbackSuhas Diggavi (University of California Los Angeles, USA)

24


Feedback is beneficial for managing bursty interference in the physical layer, wherethe burstiness of interference is caused by the lack of coordination in the upper lay-ers. In our previous work, we investigated a two-user bursty interference channel,where the presence of interference on the two interfering links is governed by a sin-gle Bernoulli random state, i.e., they are on and off simultaneously. In this work weextend our results to the case where the two interfering links can be on and off sep-arately, governed by two different Bernoulli states. When the marginal probabilitydistributions of the two states are the same, we recover our previous results. It turnsout that the capacity characterization does not depend on the joint distributions ofthe two states.(Joint work with I-Hsiang Wang)

Bio: Suhas N. Diggavi received the B. Tech. degree in electrical engineering fromthe Indian Institute of Technology, Delhi, India, and the Ph. D. degree in electricalengineering from Stanford University, Stanford, CA, in 1998. After completinghis Ph. D., he was a Principal Member Technical Staff in the Information SciencesCenter, AT&T Shannon Laboratories, Florham Park, NJ. After that he was on thefaculty of the School of Computer and Communication Sciences, EPFL, where hedirected the Laboratory for Information and Communication Systems (LICOS). Heis currently a Professor, in the Department of Electrical Engineering, at the Uni-versity of California, Los Angeles, where he directs the Information Theory andSystems laboratory. His research interests include wireless network informationtheory, wireless networking systems, network data compression and network al-gorithms. He is a co-recipient of the 2013 IEEE Information Theory Society &Communications Society Joint Paper Award, the 2013 ACM International Sympo-sium on Mobile Ad Hoc Networking and Computing (MobiHoc) best paper award,the 2006 IEEE Donald Fink prize paper award, 2005 IEEE Vehicular TechnologyConference best paper award, the Okawa foundation research award and is a Fellowof the IEEE. He has served on the editorial board for Transactions on InformationTheory, ACM/IEEE Transactions on Networking and IEEE. IEEE Communica-tion Letters, a guest editor for IEEE Selected Topics in Signal Processing and wastheTechnical Program Co-Chair for 2012 IEEE Information Theory Workshop. Hehas 8 issued patents.

Tuesday, 15:10–16:30


TuE1: Spatial Coupling

Chair: Krishna Narayanan (Texas A&M University, USA)

15:10

Thresholds of Spatially Coupled Systems via Lyapunov’sMethodChristian Schlegel (University of Alberta, Canada); Marat V. Burnashev(Institute for Information Transmission Problems, Russian Academy ofSciences, Russia)

The threshold, or saturation phenomenon of spatially coupled systems is revisitedin the light of Lyapunov’s theory of dynamical systems. It is shown that an ap-plication of Lyapunov’s direct method can be used to quantitatively describe thethreshold phenomenon, prove convergence, and compute threshold values. Thisprovides a general proof methodology for the various systems recently studied.

15:30

Displacement Convexity – A Useful Framework for the Studyof Spatially Coupled CodesRafah El-Khatib (EPFL, Switzerland); Nicolas Macris (EPFL, Switzer-land); Rudiger Urbanke (EPFL, Switzerland)

Spatial coupling has recently emerged as a powerful paradigm to construct graphi-cal models that work well under low-complexity message-passing algorithms. Al-though much progress has been made on the analysis of spatially coupled modelsunder message passing, there is still room for improvement, both in terms of sim-plifying existing proofs as well as in terms of proving additional properties.We introduce one further tool for the analysis, namely the concept of displacementconvexity. This concept plays a crucial role in the theory of optimal transport and itis also well suited for the analysis of spatially coupled systems. In cases where the

concept applies, displacement convexity allows functionals of distributions whichare not convex to be represented in an alternative form, so that they are convexwith respect to the new parametrization. The alternative convex structure can thenoften be used to prove the uniqueness of the minimizer of this functional. As aproof of concept we consider spatially coupled (l, r)-regular Gallager ensembleswhen transmission takes place over the binary erasure channel. In particular, wefirst show the existence of an optimal profile which minimizes the potential func-tional governing this system. This profile characterizes the “decoding wave” ofthe spatially coupled system. We then show that the potential function of the cou-pled system is displacement convex. Due to some translational degrees of freedomthe convexity by itself falls short of establishing the uniqueness of the minimizingprofile. But as we will discuss it is an important step in this direction.

15:50

A Closed-Form Scaling Law for Convolutional LDPC CodesOver the BECPablo M. Olmos (Universidad Carlos III de Madrid, Spain); Rudiger Ur-banke (EPFL, Switzerland)

We propose a scaling law for the error probability of convolutional LDPC ensem-bles when transmission takes place over the binary erasure channel. We discusshow the parameters of the scaling law are connected to fundamental quantitiesappearing in the asymptotic analysis of these ensembles and we verify that the pre-dictions of the scaling law fit well with data derived from simulations over a widerange of parameters.

16:10

A Finite Length Performance Analysis of LDPC Codes Con-structed by Connecting Spatially Coupled ChainsPablo M. Olmos (Universidad Carlos III de Madrid, Spain); David G. M.Mitchell (University of Notre Dame, USA); Dmitri Truhachev (Universityof Alberta, Canada); Daniel J. Costello, Jr. (University of Notre Dame,USA)

The finite length performance of codes on graphs constructed by connecting spa-tially coupled low-density parity-check (SC-LDPC) code chains is analyzed. Suc-cessive (peeling) decoding is considered for the binary erasure channel (BEC). Theevolution of the undecoded portion of the bipartite graph remaining after each it-eration is analyzed as a dynamical system. It is shown that, in addition to superioriterative decoding thresholds, connected chain ensembles have better performancethan single chain ensembles of the same rate and length.


TuE2: Ranking and Group Testing

Chair: Suhas Diggavi (University of California Los Angeles, USA)

15:10

Aggregating Rankings with Positional ConstraintsFarzad Farnoud (Hassanzadeh) (University of Illinois, Urbana-Champaign, USA); Olgica Milenkovic (University of Illinois, USA)

We consider the problem of rank aggregation, where the goal is to assemble orderedlists into one consensus order. Our contributions consist of proposing a new familyof distance measures that allow for incorporating practical ranking constraints intothe aggregation problem formulation; showing how such distance measures arisefrom a generalization of Kemeny’s axioms of the Kendall τ distance; and provingthat special classes of the proposed distances may be computed in polynomial time.

15:30

Iterative Similarity Inference via Message Passing in FactorGraphs for Collaborative FilteringJun Zou (Georgia Institute of Technology, USA); Arash Einolghozati(Georgia Tech, USA); Erman Ayday (EPFL, Switzerland); FaramarzFekri (Georgia Institute of Technology, USA)

In this paper, we develop a Belief Propagation (BP) algorithm for similarity com-putation to improve the recommendation accuracy of the neighborhood method,

25


which is one of the most popular Collaborative Filtering (CF) recommendationalgorithms. We formulate a probabilistic inference problem as to compute themarginal posterior distributions of similarity variables from their joint posteriordistribution given the observed ratings. However, direct computation is prohibitivein large-scale recommender systems. Therefore, we introduce an appropriate cho-sen factor graph to express the factorization of the joint distribution function, andutilize the BP algorithm that operates in the factor graph to exploit the factorizationfor efficient inference. In addition, since the high degree at the factor node incurs anexponential increase in computational complexity, we also propose a complexity-reduction technique. The overall complexity of the proposed BP algorithm on afactor graph is linear in the number of variables, which ensures scalability. Finally,through experiments on the MovieLens dataset, we show the superior predictionaccuracy of the proposed BP-based similarity computation algorithm for recom-mendation.

15:50

Stochastic Threshold Group TestingChun Lam Chan (The Chinese University of Hong Kong, Hong Kong);Sheng Cai (The Chinese University of Hong Kong, Hong Kong); MayankBakshi (The Chinese University of Hong Kong, Hong Kong); Sid-harth Jaggi (Chinese University of Hong Kong, Hong Kong); VenkateshSaligrama (Boston University, USA)

We formulate and analyze a stochastic threshold group testing problem motivatedby biological applications. Here a set of n items contains a subset of d ndefective items. Subsets (pools) of the n items are tested. The test outcomesare negative if the number of defectives in a pool is no larger than l; positive ifthe pool contains more than u defectives, and stochastic (negative/positive withsome probability) if the number of defectives in the pool is in the interval [l, u].The goal of a stochastic threshold group testing scheme is to identify the set ofd defective items via a “small” number of such tests with high probability. Inthe regime that l = o(d) we present schemes that are computationally feasible todesign and implement, and require near-optimal number of tests. Our schemes arerobust to a variety of models for probabilistic threshold group testing.

16:10

Optimal Binary Measurement Matrices for Compressed Sens-ingArash Saber Tehrani (University of Southern California, USA); Alexan-dros Dimakis (University of Texas at Austin, USA); Giuseppe Caire (Uni-versity of Southern California, USA)

We explicitly construct binary measurement matrices with good sparse approxi-mation guarantees. Specifically, our measurement matrices have an order optimalnumber of measurements and have `1/`1 approximation guarantee. Our construc-tion uses the progressive edge growth technique. We apply coding theoretic resultsand rely on a recent connection of compressed sensing to LP relaxation for channeldecoding.


TuE3: Multi-terminal Communication

Chair: Shlomo (Shitz) Shamai (The Technion, Israel)

15:10

Fundamental Limits of Distributed Caching in D2D WirelessNetworksMingyue Ji (University of Southern California, USA); Giuseppe Caire(University of Southern California, USA); Andreas Molisch (Universityof Southern California, USA)

We consider a wireless Device-to-Device (D2D) network where communication isrestricted to be single-hop, users make arbitrary requests from a finite library ofpossible files and user devices cache information in the form of carefully designedsets of packets from all files in the library. We consider the combined effect ofcoding in the delivery phase, achieving “coded multicast gain”, and of spatial reusedue to local short-range D2D communication. Somewhat counterintuitively, weshow that the coded multicast gain and the spatial reuse gain do not cumulate, interms of the throughput scaling laws. In particular, the spatial reuse gain shown in

our previous work on uncoded random caching and the coded multicast gain shownin this paper yield the same scaling laws behavior, but no further scaling law gaincan be achieved by using both coded caching and D2D spatial reuse.

15:30

Analog Index Coding over Block-Fading MISO BroadcastChannels with FeedbackPablo Piantanida (Supelec, France); Mari Kobayashi (Supelec, France);Giuseppe Caire (University of Southern California, USA)

We define an “analog” index coding problem where a transmitter wishes to senda set of analog sources to be reconstructed with a desired distortion level at K re-ceivers, each of which is characterized by a set of desired sources and by its ownside information. The transmission channel is a multi-input single-output (MISO)broadcast channel with independent fading, perfect channel state information atthe receiver and (strictly causal) feedback at the transmitter. An outer bound onthe rate-distortion region is derived, revealing an interesting tradeoff between thecommunication rate (source samples per channel use) and the distortion exponen-tial decay with SNR in dB. We focus on the high-SNR and low-distortion regime,in which the proposed outer bound is shown to be tight for some cases of particularinterest.

15:50

Equivalence of Inner Regions for Broadcast Channel CodingJun Muramatsu (NTT Corporation, Japan)

The aim of this paper is to investigate relationship between inner regions for broad-cast channel codes transmitting common and private messages. One of the regionsis known as the Marton inner region, another region was derived by Gel’fand andPinsker, and yet another is obtained from the saturation property and collision-resistance property. Although at first glance the Marton inner region appears to belarger than the other regions, they are in fact equivalent.

16:10

A General Framework for Statistically Characterizing the Dy-namics of MIMO ChannelsFrancisco Javier Lopez-Martinez (Stanford University, USA); EduardoMartos-Naya (University of Malaga, Spain); Jose Francisco Paris (Uni-versity of Malaga, Spain); Andrea Goldsmith (Stanford University, USA)

In multiple-input multiple-output (MIMO) systems the communication channelcan be split into s parallel single-input single-output eigenchannels. The channelgains associated with these eigenchannels depend on the magnitude of the eigen-values of the complex random matrix that characterizes the MIMO channel. Wepresent a general framework for the characterization of the dynamics of MIMOchannels. In addition, we provide new analytical results for the distribution ofthe largest eigenvalue of two correlated Wishart matrices, which enable a directevaluation of different system performance metrics such as the probability of twoconsecutive outages separated by a given time window, the level crossing rate andthe average fade duration.

Tuesday, 16:50–18:10


TuF1: LDPC Codes

Chair: Christian Schlegel (Dalhousie University, USA)

16:50

Design of Masking Matrix for QC-LDPC CodesYang Liu (Xidian University, P.R. China); Ying Li (University of Xidian,P.R. China)

The masking matrix plays an important role in constructing new classes of reg-ular and irregular quasi-cyclic low density parity check (QC-LDPC) codes. Bycoupling two identical graphs in a special way, we present a new structure of themasking matrix, whose Tanner graph can be seen as a protograph. From this per-spective, we propose a Gaussian Approximation algorithm for protograph-based

26


LDPC codes to analyze the asymptotic performance of this class of codes. It isshown that, although the proposed structure of the masking matrix has almost thesame convergence threshold as the conventional one produced randomly by pro-gressive edge growth (PEG) algorithm, the former converges faster than the latter.Simulation results illustrate that the QC-LDPC code constructed by the proposedstructure of the masking matrix has about 0.2 dB coding gains compared with theconventional one.

17:10

Nonbinary LDPC-Coded Differential Modulation: Perfor-mance and Decoding AlgorithmMinghua Li (Xidian University, P.R. China); Bao-Ming Bai (Xidian Uni-versity, P.R. China); Zhang Pei ying (Xidian University, P.R. China); XiaoMa (Sun Yat-sen University, P.R. China)

This paper is concerned with the design and performance of nonbinary LDPC-coded differential modulation systems. A low-complexity joint detection/decodingmethod for non-coherent demodulation is proposed, in which the hard-message-passing strategy is used for a joint factor graph. It combines trellis-based differen-tial detection aided with channel prediction and the reliability-based decoding ofnonbinary LDPC codes introduced in [1]. The Max-Log-MAP algorithm with soft-in hard-out is used for the differential detection. Simulation results show that theproposed method can offer good performances with a greatly reduced complexity.

17:30

Design of Non-Binary Quasi-Cyclic LDPC Codes by ACE Op-timizationAlex Bazarsky (Tel Aviv University, Israel); Noam Presman (Tel Aviv Uni-versity, Israel); Simon Litsyn (Tel Aviv University, Israel)

An algorithm for constructing Tanner graphs of non-binary irregular quasi-cyclicLDPC codes is introduced. It employs a new method for selection of edge labelsallowing control over the code’s non-binary ACE spectrum and resulting in lowerror-floor. The efficiency of the algorithm is demonstrated by generating goodcodes of short to moderate length over small fields, outperforming codes generatedby the known methods.

17:50

Improved decoding for binary source coding with coded sideinformationAnne Savard (CNRS/ENSEA/University Cergy-Pontoise, France); Clau-dio Weidmann (CNRS/ENSEA/University Cergy-Pontoise, France)

This paper presents a new iterative decoding algorithm for the source coding withcoded side information problem. Side information (SI) is compressed to an in-dex by a many-to-one (quantization) function. Instead of using the reconstructioncorresponding to the quantization index as a single representative SI word to aidthe main decoder, one can modify it by projecting an intermediate estimate of thesource word onto the Voronoi cell associated to the SI index. The hope is that theprojection brings the representative SI word closer to the source word, and thusaccelerates iterative decoding. Simulations using LDPC syndrome coding in themain branch and trellis-coded quantization in the SI branch show that for a fixednumber of decoder iterations, this method indeed increases the number of correctlydecoded source words. In fact, the decoding threshold is shifted, which may beattributed to a partial compensation of the suboptimality of the quantizer.


TuF2: Fundamental Limits of Learning and Inference

Chair: Marc Lelarge (INRIA and ENS, France)

16:50

An Impossibility Result for High Dimensional SupervisedLearningMohammad Hossein Rohban (Boston University, USA); Prakash Ishwar(Boston University, USA); Birant Orten (Turn, Inc., USA); William Karl(Boston University, USA); Venkatesh Saligrama (Boston University, USA)

We study high-dimensional asymptotic performance limits of binary supervisedclassification problems where the class conditional densities are Gaussian with un-known means and covariances and the number of signal dimensions scales fasterthan the number of labeled training samples. We show that the Bayes error, namelythe minimum attainable error probability with complete distributional knowledgeand equally likely classes, can be arbitrarily close to zero and yet the limiting min-imax error probability of every supervised learning algorithm is no better than arandom coin toss. In contrast to related studies where the classification difficulty(Bayes error) is made to vanish, we hold it constant when taking high-dimensionallimits. In contrast to VC-dimension based minimax lower bounds that consider theworst case error probability over all distributions that have a fixed Bayes error, ourworst case is over the family of Gaussian distributions with constant Bayes error.We also show that a nontrivial asymptotic minimax error probability can only be at-tained for parametric subsets of zero measure (in a suitable measure space). Theseresults expose the fundamental importance of prior knowledge and suggest thatunless we impose strong structural constraints, such as sparsity, on the parametricspace, supervised learning may be ineffective in high dimensional small samplesettings.

17:10

Information-theoretic Limits on the Classification of GaussianMixtures: Classification on the Grassmann ManifoldMatthew Nokleby (Duke University, USA); Robert Calderbank (DukeUniversity, USA); Miguel Rodrigues (University College London, UnitedKingdom)

Motivated by applications in high-dimensional signal processing, we derive funda-mental limits on the performance of compressive linear classifiers. By analogy withShannon theory, we define the classification capacity, which quantifies the maxi-mum number of classes that can be discriminated with low probability of error,and the diversity-discrimination tradeoff, which quantifies the tradeoff between thenumber of classes and the probability of classification error. For classification ofGaussian mixture models, we identify a duality between classification and commu-nications over non-coherent multiple- antenna channels. This duality allows us tocharacterize the classification capacity and diversity-discrimination tradeoff usingexisting results from multiple-antenna communication. We also identify the easiestpossible classification problems, which correspond to low-dimensional subspacesdrawn from an appropriate Grassmann manifold.

17:30

Asymptotically Minimax Regret by Bayes Mixtures for Non-exponential FamiliesJun’ichi Takeuchi (Kyushu University, Japan); Andrew R. Barron (YaleUniversity, USA)

We study the problems of data compression, gambling and prediction of a sequencexn = x1x2 . . . xn from an alphabet X , in terms of regret with respect to vari-ous families of probability distributions. It is known that the regret of the Bayesmixture with respect to a general exponential families asymptotically achieves theminimax value when variants of Jeffreys prior are used, under the condition thatthe maximum likelihood estimate is in the interior of the parameter space. We dis-cuss a modification of Jeffreys prior which has measure outside the given family ofdensities, to achieve minimax regret with respect to non-exponential type families,e.g. curved exponential families and mixture families. These results also providecharacterization of Rissanen’s stochastic complexity for those classes.

17:50

Asymptotically equivalent sequences of matrices and relativeentropyJesus Gutierrez-Gutierrez (CEIT and Tecnun (University of Navarra),Spain); Pedro M. Crespo (CEIT and TECNUN (University of Navarra),Spain)

In this paper we prove the asymptotic formula that has been recently used as anumerical integration method to approximate the relative entropy (or Kullback-Leibler distance) between two probability density functions with bounded supportin terms of functions of Hermitian Toeplitz matrices. To prove that asymptotic for-mula we use the Gray concept of asymptotically equivalent sequences of matrices.

27



TuF3: Multiple Access Channels

Chair: Dongning Guo (Northwestern University, USA)

16:50

Achievable Rates for Intermittent Multi-Access Communica-tionMostafa Khoshnevisan (University of Notre Dame, USA); J. NicholasLaneman (University of Notre Dame, USA)

We formulate a model for intermittent multi-access communication for two usersthat captures the bursty transmission of the codeword symbols for each user and thepossible asynchronism between the receiver and the transmitters as well as betweenthe transmitters themselves. By making different assumptions for the intermittentprocess, we specialize the system to a random access system with or without colli-sions. For each model, we characterize the performance of the system in terms ofachievable rate regions. The intermittency of the system comes with a significantcost in our achievable schemes.

17:10

Gaussian Many-Access Channels: Definition and SymmetricCapacityDongning Guo (Northwestern University, USA); Xu Chen (NorthwesternUniversity, USA)

This paper studies communication networks with a very large number of users si-multaneously communicating with an access point. A new notion of many-accesschannel (MnAC) is introduced, which is the same as a multiaccess channel exceptthat the number of users increases unboundedly with the coding block length. Un-like the conventional multiaccess channel with a fixed number of users, the jointtypicality decoding technique is not directly applicable to establish the achievabil-ity of the capacity. It is shown that, as long as the number of users grows sublinearlywith the coding block length, random coding with Feinstein’s threshold decodingis sufficient to achieve the symmetric capacity of the Gaussian MnAC.

17:30

MAC Capacity Under Distributed Scheduling of MultipleUsers and Linear DecorrelationJoseph Kampeas (Ben-Gurion University of the Negev, Israel); Asaf Co-hen (Ben-Gurion University of the Negev, Israel); Omer Gurewitz (BenGurion University, Israel)

Consider the problem of a multiple-antenna Multiple-Access Channel at the limitof large number of users. Clearly, in practical scenarios, only a small subset ofthe users can be scheduled to utilize the channel simultaneously. Thus, a problemof user selection arises. Since solutions which collect Channel State Information(CSI) from all users and decide on the best subset to transmit in each slot do notscale when the number of users is large, distributed algorithms for user selectionare advantageous.In this paper, we suggest distributed user selection algorithms which select a groupof users to transmit without coordinating between all users and without all userssending CSI to the base station. These threshold-based algorithms are analyzed,and their expected capacity in the limit of large number of users is investigated.It is shown that for large number of users a distributed algorithm can achieve thesame scaling laws as the optimal centralized scheme.

17:50

On the Weakest Resource for Coordination in AV-MACs withConferencing EncodersMoritz Wiese (Technische Universitat Munchen, Germany); Holger Boche(Technical University Munich, Germany)

If the senders and the receiver of an Arbitrarily Varying Multiple-Access Channel(AV-MAC) have access to the outputs of discrete correlated memoryless sources,the same rate region is achievable as if common randomness were available. Thisreduces the necessary amount of cooperation in an AV-MAC considerably. More-over, to transmit blocklength-n words, no more than order logn source outputs arerequired.

Guided Tour of Seville’s Alcazar Palace

Tuesday, 20:45–22:00 Real Alcazar (Puerta del Leon)

Wednesday

Wednesday’s Plenary, 08:40–09:40

Wednesday, 08:40–09:40 Salon de Actos

Chair: Chair: Rudiger Urbanke (EPFL, Switzerland)

Integrating Diverse Types of Data to Search for New Thera-peutic Strategies for DiseasesErnest Fraenkel (Massachusetts Institute of Technology, USA)

Biology has been transformed by new technologies that provide detailed descrip-tions of the molecular changes that occur in diseases. However, it is difficult touse these data to reveal new therapeutic insights for several reasons. Despite theirpower, each of these methods still only captures a small fraction of the cellularresponse. Moreover, when different assays are applied to the same problem, theyprovide apparently conflicting answers. I will show that network modeling revealsthe underlying consistency of the data by identifying small, functionally coherentpathways linking the disparate observations. We have used these methods to an-alyze how oncogenic mutations alter signaling and transcription and to prioritizeexperiments aimed at discovering therapeutic targets.

Bio: Ernest Fraenkel was first introduced to computational biology in high schoolwhen the field did not yet have a name. His early experiences with Professor CyrusLevinthal of Columbia University taught him that biological insights often comefrom unexpected disciplines. After graduating summa cum laude from HarvardCollege in Chemistry and Physics he obtained his Ph. D. at MIT in the departmentof Biology and did post-doctoral work at Harvard. As the field of Systems Biol-ogy began to emerge, he established a research group in this area at the WhiteheadInstitute and then moved to the Department of Biological Engineering at the Mas-sachusetts Institute of Technology. His research group takes a multi-disciplinaryapproach involving tightly connected computational and experimental methods touncover the molecular pathways that are altered in cancer, neurodegenerative dis-eases, and diabetes.

Wednesday, 09:50–11:10

Wednesday, 09:50–11:10 Room 1 – Sala Juan Larraneta

WeB1: Real Number Codes

Chair: Jean-Claude Belfiore (Ecole Nationale Superieure desTlcommunications, France)

09:50

A new design criterion for spherically-shaped divisionalgebra-based space-time codesLaura Luzzi (ENSEA, France); Roope Vehkalahti (University of Turku,Finland)

This work considers normalized inverse determinant sums as a tool for analyz-ing the performance of division algebra based space-time codes for multiple an-tenna wireless systems. A general union bound based code design criterion isobtained as a main result. In our previous work, the behavior of inverse deter-minant sums was analyzed using point counting techniques for Lie groups; it wasshown that the asymptotic growth exponents of these sums correctly describe thediversity-multiplexing gain trade-off of the space-time code for some multiplexing

28


gain ranges. This paper focuses on the constant terms of the inverse determinantsums, which capture the coding gain behavior. Pursuing the Lie group approach,a tighter asymptotic bound is derived, allowing to compute the constant terms forseveral classes of space-time codes appearing in the literature. The resulting designcriterion suggests that the performance of division algebra based codes depends onseveral fundamental algebraic invariants of the underlying algebra.

10:10

Projections, Dissections and Bandwidth Expansion MappingsAntonio Campello (University of Campinas, Brazil); Vinay A. Vaisham-payan (AT&T Labs - Research, USA); Sueli I. R. Costa (State Universityof Campinas-UNICAMP, Brazil)

We address the problem of constructing explicit mappings from a k-dimensionalcontinuous alphabet source to an n-dimensional Gaussian channel. The source isassumed to be uniformly distributed on the unit cube [0, 1)k . The scheme con-sidered is based on a family of piecewise linear mappings and its performance isshown to be related to specific projected lattices of Zn. We study sufficient con-ditions for the mean squared error of such mappings to scale optimally with thesignal-to-noise ratio of the channel and present an explicit construction for the casek = n − 1. However, in some other cases our scheme requires the source to beuniformly distributed over a fundamental region of a specific lattice, that may benot congruent to [0, 1)k . A dissection technique is presented in order to overcomethe source support mismatch and the MSE degradation of such a transformation isanalyzed. An example construction of a 2 : n expansion mapping using the dis-section technique is presented and is shown to exhibit optimal scaling of the MSEwith the channel SNR.

10:30

Robust error correction for real-valued signals via message-passing decoding and spatial couplingJean Barbier (ESPCI, France); Florent Krzakala (ESPCI, France); LenkaZdeborova (Institut de Physique Theorique IPhT, CEA Saclay and CNRS,France); Pan Zhang (ESPCI, France)

We revisit the error correction scheme of real-valued signals when the codewordis corrupted by gross errors on a fraction of entries and a small noise on all theentries. Combining the recent developments of approximate message passing andthe spatially-coupled measurement matrix in compressed sensing we show that theerror correction and its robustness towards noise can be enhanced considerably.We discuss the performance in the large signal limit using previous results on stateevolution, as well as for finite size signals through numerical simulations. Even forrelatively small sizes, the approach proposed here outperforms convex-relaxation-based decoders.

10:50

Packing Tubes on Tori: An Efficient Method for Low SNRAnalog Error CorrectionRobert M. Taylor, Jr. (MITRE Corporation, USA); Lamine Mili (VirginiaTech, USA); Amir Zaghloul (Virginia Polytechnic Institute and State Uni-versity, USA)

In this study we introduce a new class of bandwidth-expansion source-channelcodes for analog error correction that can work in any even dimensional spaceand show superior performance in the low SNR region. Our codes are constructedas geodesics on flat tori as previous studies have done, but we use a tube radiusconstruct derived from the global circumradius function. We prove that geodesicson flat tori lead to constant generalized curvature curves and exploit that propertyto give a single-argument form for the circumradius function. We optimize encoderparameters by minimizing the global radius of curvature subject to harmonic fre-quency structure which leads to closed twisted tubes with maximal tube radius onthe tori. We exploit the isometry of the flat torus with the hyperrectangle to derivesimple closed-form decoders based on torus projections that come with 2 dB of themaximum likelihood decoder.

Wednesday, 09:50–11:10 Room 2 – Sala de Grados

WeB2: Inference, Information and Complexity

Chair: Ioannis Kontoyiannis (Athens UniversityEcon & Business,Greece)

09:50

Reconstruction in the Labeled Stochastic Block ModelMarc Lelarge (INRIA and ENS, France); Laurent Massoulie (MicrosoftResearch - INRIA Joint Center, Sweden); Jiaming Xu (University of Illi-nois at Urbana-Champaign, USA)

The labeled stochastic block model is a random graph model representing networkswith community structure and interactions of multiple types. In its simplest form, itconsists of two communities of approximately equal size, and the edges are drawnand labeled at random with probability depending on whether their two endpointsbelong to the same community or not.It has been conjectured in [1] that this model exhibits a phase transition: recon-struction (i.e. identification of a partition positively correlated with the “true par-tition” into the underlying communities) would be feasible if and only if a modelparameter exceeds a threshold.We prove one half of this conjecture, i.e., reconstruction is impossible when belowthe threshold. In the converse direction, we introduce a suitably weighted graph.We show that when above the threshold by a specific constant, reconstruction isachieved by (1) minimum bisection, and (2) a spectral method combined with re-moval of nodes of high degree.

10:10

Information-Preserving Markov AggregationBernhard C. Geiger (Graz University of Technology, Austria); ChristophTemmel (VU University Amsterdam, The Netherlands)

We present a sufficient condition for a non-injective function of a Markov chain tobe a second-order Markov chain with the same entropy rate as the original chain.This permits an information-preserving state space reduction by merging statesor, equivalently, lossless compression of a Markov source on a sample-by-samplebasis. The cardinality of the reduced state space is bounded from below by thenode degrees of the transition graph associated with the original Markov chain.We also present an algorithm listing all possible information- preserving state spacereductions, for a given transition graph. We illustrate our results by applying thealgorithm to a bi-gram letter model of an English text.

10:30

On the Noise Sensitivity and Mutual Information of (Nested-)Canalizing Boolean FunctionsJohannes Georg Klotz (Ulm University, Germany); Martin Bossert (UlmUniversity, Germany); Steffen Schober (Ulm University, Germany)

We investigate the mutual information of Boolean functions with noisy inputs.Therefore, we derive a relation between the noise sensitivity and the mutual in-formation. Further, we apply Fourier analysis to give upper bounds on the noisesensitivity and lower bounds on the mutual information for canalizing and nestedcanalizing functions. From these bounds we conjecture the optimality of theseclasses of functions.

10:50

Coupled Neural Associative MemoriesAmin Karbasi (ETHZ, Switzerland); Amir Hesam Salavati (Ecole Poly-technique Federale de Lausanne, Switzerland); Amin Shokrollahi (EPFL,Switzerland)

We propose a novel architecture to design a neural associative memory that is ca-pable of learning a large number of patterns and recalling them later in presenceof noise. It is based on dividing the neurons into local clusters and parallel plains,an architecture that is similar to the visual cortex of macaque brain. The commonfeatures of our proposed model with those of spatially-coupled codes enable usto show that the performance of such networks in eliminating noise is drasticallybetter than the previous approaches while maintaining the ability of learning an ex-ponentially large number of patterns. Previous work either failed in providing good

29


performance during the recall phase or in offering large pattern retrieval (storage)capacities. We also present computational experiments that lend additional supportto the theoretical analysis.

Wednesday, 09:50–11:10 Room 3 – Aula 304

WeB3: Network Coding

Chair: Terence H. Chan (University of South Australia, Australia)

09:50

Linear Network Coding for Multiple Groupcast Sessions: AnInterference Alignment ApproachAbhik Kumar Das (The University of Texas at Austin, USA); SiddharthaBanerjee (The University of Texas at Austin, USA); Sriram Vishwanath(University of Texas Austin, USA)

We consider the problem of linear network coding over communication networks,representable by directed acyclic graphs, with multiple groupcast sessions: the net-work comprises of multiple destination nodes, each desiring messages from mul-tiple sources. We adopt an interference alignment perspective, providing new in-sights into designing practical network coding schemes as well as the impact ofnetwork topology on the complexity of the alignment scheme. In particular, weshow that under certain (polynomial-time checkable) constraints on networks withK sources, it is possible to achieve a rate of 1/(L+ d+ 1) per source using linearnetwork coding coupled with interference alignment, where each destination re-ceives messages from L sources (0 < K), and d is a parameter, solely dependenton the network topology, that satisfies 0 ≤ d < K − L.

10:10

On the capacity of ms/3t and 3s/nt sum-networksBrijesh Kumar Rai (IIT Guwahati, India); Niladri Das (IIT Guwahati,India)

We consider directed acyclic networks where each terminal requires sum of allthe sources. Such a class of networks has been termed as sum-networks in theliterature. A sum-network having m sources and n terminals has been termed asa ms/nt sum-network. There has been previous works on the capacity of sum-networks, specifically, it has been shown that the capacity of a 3s/3t sum-networkis either 0, 2/3 or ≥ 1. In this paper, we consider some generalizations of 3s/3tsum-networks, namely, ms/3t and 3s/nt sum-networks, where m,n ≥ 3. Forms/3t and 3s/nt sum-networks, where m,n ≥ 3, if the min-cut between eachsource and each terminal is at least 1, the capacity is known to be at least 2/3.In this paper, we show that there exist ms/3t and 3s/nt sum-networks whosecapacities lie between 2/3 and 1. Specifically, we show that for any positive integerk ≥ 2, there exists a ms/3t sum-network (and also a 3s/nt sum-network) whosecapacity is k

k+1. We conjecture that the capacity of a ms/3t sum-network, where

m > 3 (and also of a 3s/nt sum-network, where n > 3) is either 0,≥ 1 or of theform k

k+1, where k is a positive integer greater than or equal to 2.

10:30

Combinatorial Flow over Cyclic Linear NetworksChung Chan (The Chinese University of Hong Kong, Hong Kong); Ken-neth W. Shum (Institute of Network Coding, Hong Kong); Qifu T Sun (TheChinese University of Hong Kong, Hong Kong)

A combinatorial notion of flow is identified for time-invariant linear coding overnon-layered deterministic linear networks that may contain cycles, broadcast andinterference links. It reveals the matroidal structure for efficient code construction,and enables a seamless extension of the classical network coding results. In addi-tion to a max-flow min-cut theorem for multicast, time-invariant routing is shownto be optimal for unicast.

10:50

Outer Bounds and a Functional Study of the Edge RemovalProblemEun Jee Lee (California Institute of Technology, USA); Michael Langberg(Open University of Israel, Israel); Michelle Effros (California Instituteof Technology, USA)

In this paper, we investigate the impact of a single edge on the capacity regionof a network of error-free, point-to-point links. A family of networks and edgesis said to exhibit the “edge removal property” if for any network and edge in thefamily, removing a δ-capacity edge changes the capacity region by at most δ ineach dimension. We derive a sufficient condition on network coding functions toguarantee that the edge removal property holds when the network is operated usingfunctions satisfying the condition. Also, we extend the family of network capacitybounds for which it is known that removing a single edge of capacity δ changesthe capacity bound by at most f(δ) in each dimension. Specifically, we show thatremoving a single δ-capacity edge changes the Generalized Network Sharing outerbound by at most δ in each dimension and the Linear Programming outer bound byat most a constant times δ in each dimension.

Wednesday, 11:30–12:50

Wednesday, 11:30–12:50 Room 1 – Sala Juan Larraneta

WeC1: Lattice Codes

Chair: Ramji Venkataramanan (University of Cambridge, United King-dom)

11:30

Lattice Codes Based on Product Constructions over F2q with

Applications to Compute-and-ForwardYu-Chih Huang (Texas A&M University, USA); Krishna Narayanan(Texas A&M University, USA)

A novel construction of lattices is proposed. This construction can be thought of asConstruction A with linear codes that can be represented as the Cartesian productof two linear codes over Fq ; hence, is referred to as the product construction. Theexistence of a sequence of Poltyrev-good lattices generated by the product con-struction under some conditions is shown. This family of lattices is then used togenerate signal constellations with q2 elements which can be used in conjunctionwith multilevel coding with channel codes over Fq instead of Fq2 to design goodcoded modulation schemes for compute-and-forward.

11:50

Precoded Integer-Forcing Universally Achieves the MIMOCapacity to Within a Constant GapOr Ordentlich (Tel Aviv University, Israel); Uri Erez (Tel Aviv University,Israel)

An open-loop single-user multiple-input multiple-output communication scheme isconsidered where a transmitter, equipped with multiple antennas, encodes the datainto independent streams all taken from the same linear code. The coded streamsare then linearly precoded using the encoding matrix of a perfect linear dispersionspace-time code. At the receiver side, integer-forcing equalization is applied, fol-lowed by standard single-stream decoding. It is shown that this communicationarchitecture achieves the capacity of any Gaussian multiple-input multiple-outputchannel up to a gap that depends only on the number of transmit antennas.

12:10

Enabling Multiplication in Lattice Codes via Construction AFrederique Oggier (Nanyang Technological University, Singapore); Jean-Claude Belfiore (Ecole Nationale Superieure des Telecommunications,France)

As a first step towards distributed computations in a wireless network, we introduceideal lattices, that is lattices built over an ideal of a ring of integers in a number field,as a tool for constructing lattice codes at the physical layer. These lattices are notonly additive groups as all lattices, but they are also equipped with a multiplication,which enables polynomial operations at each node of the wireless network. In thispaper, we show how some of these ideal lattices can be constructed from polyno-mial codes (generalization of cyclic codes) via Construction A, and illustrate howthese lattices enable multiplication.

12:30

Spatially-Coupled Low Density Lattices based on Construc-

30


tion A with Applications to Compute-and-ForwardNihat Tunali (Texas A&M University, USA); Krishna Narayanan (TexasA&M University, USA); Henry D. Pfister (Texas A&M University, USA)

We consider a class of lattices built using Construction A, where the underlyingcode is a non-binary spatially-coupled low density parity check code. We referto these lattices as spatially-coupled LDA (SCLDA) lattices. SCLDA lattices canbe constructed over integers, Gaussian integers and Eisenstein integers. We em-pirically study the performance of SCLDA lattices under belief propagation (BP)decoding. Ignoring the rate loss from termination, simulation results show that theBP thresholds of SCLDA lattices over integers is 0.11 dB (0.34 dB with the rateloss) and the BP thresholds for SCLDA lattices over Eisenstein integers are 0.08dB from the Poltyrev limit (0.19 dB with the rate loss). Motivated by this result,we use SCLDA lattice codes over Eisenstein integers for implementing a compute-and-forward protocol. For the examples considered in this paper, the thresholdsfor the proposed lattice codes are within 0.28 dB from the achievable rate of thiscoding scheme and within 1.06 dB from the achievable computation rate of Nazerand Gastpar’s coding scheme in [6] extended to Eisenstein integers.

Wednesday, 11:30–12:50 Room 2 – Sala de Grados

WeC2: Entropy and Statistics of Sequences

Chair: Neri Merhav (Technion, Israel)

11:30

Information Theory for Atypical SequencesAnders Høst-Madsen (University of Hawaii, USA); Elyas Sabeti (Univer-sity of Hawaii, USA); Chad Walton (University of Hawaii, USA)

One characteristic of the information age is the exponential growth of informa-tion, and the ready availability of this information through networks, including theinternet – “Big Data.” The question is what to do with this enormous amount ofinformation. One possibility is to characterize it through statistics – think averages.The perspective in this paper is the opposite, namely that most of the value in theinformation is in the parts that deviate from the average, that are unusual, atypi-cal. Think of art: the valuable paintings or writings are those that deviate fromthe norms, that are atypical. The same could be true for venture development andscientific research.The paper first discusses what exactly should be understood by “atypical.” This isby no means straightforward. It has to be a well defined theoretical concept cor-responding to some intuitive idea of atypicality, which when applied gives usefulresults. This is followed by a simple example of iid binary sequences. This exampleis simple enough that complete algorithms can be developed and analyzed, whichgive insights into atypicality. We finally develop a more general algorithm basedon the Context Tree Weighing algorithm and apply that to heart rate variability.

11:50

A Phase Transition for the Uniform Distribution in the PatternMaximum Likelihood ProblemWinston Fernandes (Indian Institute of Science, India); Navin Kashyap(Indian Institute of Science, India)

In this paper, we consider the setting of the pattern maximum likelihood (PML)problem studied by Orlitsky et al. We present a well-motivated heuristic algorithmfor deciding the question of when the PML distribution of a given pattern is uni-form. The algorithm is based on the concept of a “uniform threshold”. This is athreshold at which the uniform distribution exhibits an interesting phase transitionin the PML problem, going from being a local maximum to being a local minimum.

12:10

Shannon Entropy Estimation from Convergence Results in theCountable Alphabet CaseJorge F. Silva (University of Chile, Chile); Patricio Parada (Universidadde Chile, Chile)

In this paper new results for the Shannon entropy estimation and estimation of dis-tributions, consistently in information divergence, are presented in the countablealphabet case. Sufficient conditions for the entropy convergence are adopted, in-cluding scenarios with both finitely and infinitely supported distributions. From

this approach, new estimates, strong consistency results and rate of convergencesare derived for various plug-in histogram-based schemes.

12:30

The Entropy of Sums and Rusza’s Divergence on AbelianGroupsIoannis Kontoyiannis (Athens UniversityEcon & Business, Greece); Mok-shay Madiman (University of Delaware, USA)

Motivated by a series of recently discovered inequalities for the sum and differenceof discrete or continuous random variables (by Tao and the authors), we argue thatthe most natural, general form of these results is in terms of a special case of amutual information, which we call the Ruzsa divergence between two probabilitydistributions. This can be defined for arbitrary pairs of random variables takingvalues in any discrete (countable) set, on n-dimensional Euclidean space, or in facton any locally compact Hausdorff abelian group. We study the basic properties ofthe Rusza divergence and derive numerous consequences. In particular, we showthat all of the inequalities mentioned can be stated and proved in a unified way,extending their validity to the present general setting. For example, consequencesof the basic properties of the Ruzsa divergence developed here include the fact thatthe entropies of the sum and the difference of two independent random vectorsseverely constrain each other, as well as entropy analogues of a number of resultsin additive combinatorics. Although the setting is quite general, the results arealready of interest (and new) in the case of real-valued random vectors. For in-stance, another consequence is an entropic analogue (in the setting of log-concavedistributions) of the Rogers-Shephard inequality for convex bodies.

Wednesday, 11:30–12:50 Room 3 – Aula 304

WeC3: Cooperation and Interference

Chair: Pablo Piantanida (Supelec, France)

11:30

Cooperative Energy Harvesting Communications with Relay-ing and Energy SharingKaya Tutuncuoglu (The Pennsylvania State University, USA); Aylin Yener(Pennsylvania State University, USA)

This paper considers two-hop communication networks where the transmitters har-vest their energy in an intermittent fashion. In this network, communication is car-ried out by signal cooperation, i.e., relaying. Additionally, the transmitters havethe option of transferring energy to one another, i.e., energy cooperation. Energy ispartially lost during transfer, exposing a trade-off between energy cooperation anduse of harvested energy for transmission. A multi-access relay model is consideredand transmit power allocation and energy transfer policies that jointly maximizethe sum-rate are found. It is shown that a class of power policies achieves the opti-mal sum-rate, allowing a separation of optimal energy transfer and optimal powerallocation problems. The optimal energy transfer policy is shown to be an orderednode selection, where nodes with better energy transfer efficiency and worse chan-nels transfer all their energy to the relay or other source nodes via the relay. For thespecial case of single source, the optimal policy requires the direction of energytransfer to remain unchanged unless either node depletes all of its energy. Overall,the findings provide the insight that cooperation of the source nodes by sharing en-ergy with the relay node leads to them indirectly cooperating with each other, andthat such cooperation can be carried out in a last-minute fashion.

11:50

Zero vs. ε Error in Interference ChannelsIlia Levi (The Open University of Israel, Israel); Danny Vilenchik (Weiz-mann Institute of Science, Israel); Michael Langberg (Open University ofIsrael, Israel); Michelle Effros (California Institute of Technology, USA)

Traditional studies of multi-source, multi-terminal interference channels typicallyallow a vanishing probability of error in communication. Motivated by the study ofnetwork coding, this work addresses the task of quantifying the loss in rate wheninsisting on zero error communication in the context of interference channels.

12:10

The Sum-Capacity of different K-user Cognitive Interference

31


Channels in Strong InterferenceDiana Maamari (University of Illinois At Chicago, USA); Natasha De-vroye (University of Illinois at Chicago, USA); Daniela Tuninetti (Uni-versity of Illinois at Chicago, USA)

This work considers differentK-user extensions of the two-user cognitive interfer-ence channel model. The models differ by the cognitive abilities of the transmitters.In particular, the primary message sharing model, in which only one user is cogni-tive and knows all messages, and the cumulative message sharing model, in whicha user knows the messages of all users with lesser index, are analyzed.The central contribution is the characterization of the sum-capacity of both modelsunder a strong interference condition, which amounts to having one receiver in thenetwork that can decode all transmitted signals without loss of optimality. Thesum-capacity is evaluated for the Gaussian noise channel, as well as the conditionson the channel gains that grant strong interference.

12:30

On Degrees-of-Freedom of Full-Duplex Uplink/DownlinkChannelAchaleshwar Sahai (Rice University, USA); Suhas Diggavi (Universityof California Los Angeles, USA); Ashutosh Sabharwal (Rice University,USA)

Feasibility of full-duplex opens up the possibility of applying it to cellular net-works to operate uplink and downlink simultaneously for multiple users. However,simultaneous operation of uplink and downlink poses a new challenge of intra-cellinter-node interference. In this paper, we identify scenarios where inter-node inter-ference can be managed to provide significant gain in degrees of freedom over theconventional half-duplex cellular design.

Wednesday’s Plenaritas, 14:00–15:30

Wednesday, 14:00–15:30 Salon de Actos

Chair: Juan Jose Murillo-Fuentes (Universidad de Sevilla, Spain)

14:00

Information and Control in Sensing-Acting systems: Predic-tive Information and the Emergence of Hierarchical Repre-sentationsNaftali Tishby (The Hebrew University, Israel)

A major theoretical challenge for understanding intelligent systems is the connec-tion between control and information processing. While it is clear that optimal con-trol theory utilizes sensory information for state estimation, optimal control fails toproperly describe active sensing, exploration and learning, in which informationgains – rather than physical value – should be part of the utility function. I willdescribe a new theoretical framework for sensing-acting systems which correctlycombines information gathering, learning, and control. It extends the standard op-timal control paradigm (AKA reinforcement learning) by including informationconstraints on sensory, memory, and control channels. We achieve this by apply-ing large deviation bounds to the POMDP formulation of optimal control, and showthat a proper formulation of the problem should balance between “value to-go” and“information to-go” in way that precisely incorporates the capacity and actionableinformation constraints. I then argue that when there is learning, the “predictiveinformation” – information between the past and future of the process – becomessub-extensive (grows sub linearly with time) and the above principle must be mod-ified to remain asymptotically consistent. One intriguing possible modificationis the rescaling/renormalization of time with information. I will show that suchtime-renormaization of the optimal control (Bellman) equations can explain thediscounting of rewards, as well as the emergence of hierarchical representations inboth sensory perception and planning.

Bio: Naftali Tishby is the Flinkman Professor of brain research and the director ofthe Interdisciplinary Center for Neural Computation at the Hebrew University ofJerusalem. Tishby is a theoretical physicist by training, working at the interfacebetween statistical physics, information theory, and machine learning. He was aresearcher and professor at MIT, Bell Labs, Princeton NECI and UPenn, among

other institutions. Since 1992 he is a faculty of computer science and the headof the Machine Learning lab at the Hebrew university, and is well known for hisworks on the information theoretic foundations of learning and dynamical systems.

14:30

Source Coding, Lists, and Renyi EntropyAmos Lapidoth (ETHZ, Switzerland)

A sequence produced by a memoryless source is to be described using a fixed num-ber of bits that is proportional to its length. Based on the description, a list that isguaranteed to contain the sequence must be produced. The trade-off between thedescription length and the moments of the listsize is studied when the sequence’slength tends to infinity. It is characterized by the source’s Renyi entropy. Exten-sions to scenarios with side information are also studied, where the key is condi-tional Renyi entropy. The lossy case where at least one of the elements of the listmust be within a specified distortion from the source sequence is also solved.

(Joint work with Christoph Bunte)

Bio: Amos Lapidoth received the B. A. degree in mathematics (1986), the B.Sc.degree in electrical engineering (1986), and the M.Sc. degree in electrical engi-neering (1990) all from the Technion-Israel Institute of Technology. He receivedthe Ph. D. degree in electrical engineering from Stanford University in 1995. Inthe years 1995-1999 he was an Assistant and Associate Professor at the Depart-ment of Electrical Engineering and Computer Science at the Massachusetts Insti-tute of Technology, and was the KDD Career Development Associate Professor inCommunications and Technology. He is now Professor of Information Theory atETH Zurich in Switzerland. He is the author of the book A Foundation in DigitalCommunication, published by Cambridge University Press in 2009. His researchinterests are in digital communications and information theory.

15:00

The Role of the Hypercontractivity Ribbon in InformationTheoryVenkat Anantharam (University of California, Berkeley, USA)

A communication channel defines a linear map from functions on the output alpha-bet to those on the input alphabet given by taking the conditional expectation. Foreach q ≥ 1 this maps Lq functions to Lq functions and is a contraction. However,generally speaking, more is true, in that this map is a contraction fromLq functionsto Lp functions for a range of p > q. This property, which has been studied bymathematicians for many years in the broader context of linear operators betweenfunctions on abstract spaces, is called hypercontractivity. Versions of this theoryare also available q ∈ (0, 1), and it has also begun to be extended recently to neg-ative q, though of course Lq spaces for q < 1 behave quite differently from thosefor q ≥ 1. The set of p ≥ q ≥ 1 for which Lq functions are mapped contrac-tively to Lp functions is called the forward part of the hypercontractivity ribbonassociated to the channel (the full ribbon also has a backward part, for q ≤ p ≤ 1).The hypercontractivity ribbon is thus a characteristic of the channel. It turns outthat the hypercontractivity ribbon has deep and intimate relevance to the traditionalconcerns of information theory over channels. The talk will attempt to justify thepreceding sentence, for discrete memoryless channels. In particular, we will dis-cuss the role of the hypercontractivity ribbon in the distributed simulation of jointprobability distributions, as well as its role in strong data processing inequalities.

(Parts of this talk are based on joint work with Sudeep Kamath, Chandra Nair, andAmin Aminzadeh Gohari)

Bio: Venkat Anantharam is on the faculty of the Department of Electrical Engi-neering and Computer Science at the University of California at Berkeley (UCB).He received the B. Tech. degree in electronics in 1980 from the Indian Institute ofTechnology, Madras (IIT-M), and the M. A. and C. Phil. degrees in mathematicsand the M. S. and Ph. D. degrees in electrical engineering in 1983, 1984, 1982,and 1986 respectively, all from UCB. From 1986 to 1994, he was on the facultyof the School of Electrical Engineering at Cornell University. Dr. Anantharam re-ceived the Philips India Medal and the President of India Gold Medal from IIT-Min 1980, and an NSF Presidential Young Investigator award (1988-1993). He is aco-recipient of the 1998 Prize Paper Award of the IEEE Information Theory Soci-ety, and a co-recipient of the 2000 Stephen O.Rice Prize Paper Award of the IEEECommunications Theory Society. He received the Distinguished Alumnus Awardfrom IIT-M in 2008.

32


Thursday

Thursday’s Plenary, 08:40–09:40

Thursday, 08:40–09:40 Salon de Actos

Chair: Alfonso Martinez (Universitat Pompeu Fabra, Spain)

Growth Optimal Empirical Portfolio Selection StrategiesLaszlo Gyorfi (Budapest University of Technology and Economics, Hun-gary)

In this talk we survey discrete time, multi period, sequential investment strategiesfor financial markets. Under memoryless assumption on the underlying marketprocess of relative prices the best constantly rebalanced portfolio is studied, calledlog-optimal portfolio, which achieves the maximal asymptotic average growth rate.For generalized dynamic portfolio selection, when the market process is stationaryand ergodic, universal, growth optimal empirical methods are shown, using cur-rent principles of nonparametric regression estimation and machine learning algo-rithms. The empirical performance of the methods are illustrated for NYSE data.

Bio: Laszlo Gyorfi was born in Hercegfalva, Hungary, on July 6,1947. He receiveda degree in mathematics and physics in 1970 from the Lorand Eotvos University,Budapest, Hungary. He coauthored, with L. Devroye the book NonparametricDensity Estimation: the L1 View, (New York: Wiley, 1985), with L. Devroye andG. Lugosi the book Probabilistc Theory of Pattern Recognition, (Springer, 1996),and with M. Kohler, A. Krzyzak and H. Walk the book A Distribution-Free Theoryof Nonparametric Regression, (Springer, 2002). His main intersts are nonparamet-ric statistics and multiple access communication. He is now with the Departmentof Computer Science and Information Theory, Budapest University of Technologyand Economics, Budapest, Hungary. Dr. Gyorfi became an Ordinary Member ofthe Hungarian Academy of Sciences in 2001.

Thursday, 09:50–11:10

Thursday, 09:50–11:10 Room 1 – Sala Juan Larraneta

ThB1: Algebraic Codes

Chair: Navin Kashyap (Indian Institute of Science, India)

09:50

Computing the Camion’s multivariate BCH boundJose Joaquın Bernal (University of Murcia, Spain); Diana Bueno-Carreno (Pontificia Universidad Javeriana - Cali, Colombia); JuanSimon (Universidad de Murcia, Spain)

The P. Camion’s apparent distance of an abelian code is a generalization of thenotion of the BCH bound of cyclic codes [3]. In this work, we present a methodof computation of the apparent distance in multivariate abelian codes, based onmanipulations of hypermatrices. Our algorithm needs fewer computations than anyother, up to our knowledge; in fact, in the case of two dimensional abelian codes ithas linear complexity. We give two applications. First, we construct abelian codesthat multiply the dimension of a given cyclic code and equal its BCH bound. Thesecond one is an approximation to a notion of BCH multivariate code.

10:10

Flag Orbit Codes and Their Expansion to Stiefel CodesRenaud-Alexandre Pitaval (Aalto University, Finland); Olav Tirkkonen(Aalto University, Finland)

We discuss group orbits codes in homogeneous spaces for the unitary group, knownas flag manifolds. The distances used to describe the codes arise from embeddingthe flag manifolds into Euclidean hyperspheres, providing a generalization of thespherical embedding of Grassmann manifolds equipped with the so-called chordal

distance. Flag orbits are constructed by acting with a unitary representation of afinite group. In the construction, the center of the finite group has no effect, andthus it is sufficient to consider its inner automorphism group. Accordingly, someexplicit constructions from projective unitary representations of finite groups in 2and 4 dimensions are described. We conclude with examples of codes on the Stiefelmanifold constructed as orbits of the linear representation of the projective groups,and thus expansion of the flag codes considered.

10:30

New Geometrical Spectra of Linear Codes with Applicationsto Performance AnalysisXiao Ma (Sun Yat-sen University, P.R. China); Jia Liu (Sun Yat-sen Uni-versity, P.R. China); Qiutao Zhuang (Sun Yat-sen University, P.R. China);Bao-Ming Bai (Xidian University, P.R. China)

In this paper, new enumerating functions for linear codes are defined, includingthe triangle enumerating function and the tetrahedron enumerating function, bothof which can be computed using a trellis-based algorithm over polynomial rings.The computational complexity is dominated by the complexity of the trellis. Inaddition, we show that these new enumerating functions can be used to improveexisting performance bounds on the maximum likelihood decoding.

10:50

Concatenated Permutation Block Codes based on Set Parti-tioning for Substitution and Deletion Error-ControlReolyn Heymann (University of Johannesburg, South Africa); Jos H. We-ber (Delft University of Technology, The Netherlands); Theo G. Swart(University of Johannesburg, South Africa); Hendrik C. Ferreira (Uni-versity of Johannesburg, South Africa)

A new class of permutation codes is presented where, instead of considering onepermutation as a codeword, codewords consist of a sequence of permutations. Theadvantage of using permutations, i.e. their favourable symbol diversity properties,is preserved. Additionally, using sequences of permutations as codewords, coderates close to the optimum rate can be achieved. Firstly, the complete set of per-mutations is divided into subsets by using set partitioning. Binary data is thenmapped to permutations from these subsets. These permutations, together with aparity permutation, will form the codeword. Two constructions will be presented:one capable of detecting and correcting substitution errors and the other capable ofdetecting and correcting either substitution or deletion errors.

Thursday, 09:50–11:10 Room 2 – Sala de Grados

ThB2: Statistics and Learning(Organized by Gabor Lugosi)

Chair: Gabor Lugosi (Universitat Pompeu Fabra, Spain)

09:50

From Bandits to Experts: A Tale of Domination and Indepen-denceNoga Alon (Tel Aviv University, Israel); Nicolo Cesa-Bianchi (Univer-sita degli Studi di Milano, Italy); Claudio Gentile (Universita degli Studidell’Insubria, Italy); Yishay Mansour (Tel-Aviv University, Israel)

We consider the partial observability model for multi-armed bandits, introducedby Mannor and Shamir. Our main result is a characterization of regret in the di-rected observability model in terms of the dominating and independence numbersof the observability graph. We also show that in the undirected case, the learnercan achieve optimal regret without even accessing the observability graph beforeselecting an action. Both results are shown using variants of the Exp3 algorithmoperating on the observability graph in a time-efficient manner.

10:10

On Semi-Probabilistic Universal PredictionAlexander Rakhlin (University of Pennsylvania, USA); Karthik Sridharan(University of Pennsylvania, USA)

We describe two scenarios of universal prediction, as well as some recent advances

33


in the study of minimax regret and algorithmic development. We then propose anintermediate scenario, the Semi-Probabilistic Setting, and make progress towardsunderstanding the associated minimax regret.

10:30

Learning Joint Quantizers for Reconstruction and PredictionMaxim Raginsky (University of Illinois at Urbana-Champaign, USA)

We consider the problem of empirical design of variable-rate quantizers for recon-struction and prediction. When a discriminative model (conditional distribution ofthe unobserved output given the observed input) is known or can be accurately esti-mated from a separate training set, we show that this problem reduces to designinga certain type of a generalized quantizer by means of empirical risk minimizationon unlabeled input samples only. We derive a high-probability upper bound on theresulting expected performance of such a quantizer in terms of the training samplesize and the complexity parameters of the reconstruction and the prediction prob-lems. We also discuss two illustrative examples: binary classification with absoluteloss and the information bottleneck.

10:50

Prediction of Individual Sequences with Finite-Memory Uni-versal PredictorsMeir Feder (Tel-Aviv University, Israel)

Universal prediction of individual sequences is a well studied subject with applica-tions in universal coding, portfolio theory, estimation and so on. The main themeof the problem assumes a constrained, non- universal predictor that observes theentire sequence, yet since it is constrained, e.g., to be a finite-state machine, its per-formance is limited. Universal prediction theory shows that in some cases a non-anticipating universal predictor can attain this performance of the non-universalpredictor, for all individual sequences. The presented work considers the casewhere the universal predictor is also constrained to be a finite state machine.

Thursday, 09:50–11:10 Room 3 – Aula 304

ThB3: Interference and Relay Channels

Chair: Ayfer Ozgur (Stanford University, USA)

09:50

Multilevel Topological Interference ManagementChunhua Geng (University of California, Irvine, USA); Hua Sun (Univer-sity of California, Irvine, USA); Syed Ali Jafar (University of CaliforniaIrvine, USA)

The robust principles of treating interference as noise (TIN) when it is sufficientlyweak, and avoiding it when it is not, form the background for this work. Com-bining TIN with the topological interference management (TIM) framework thatidentifies optimal interference avoidance schemes, a baseline TIM-TIN approachis proposed which decomposes a network into TIN and TIM components, allocatesthe signal power levels to each user in the TIN component, allocates signal vec-tor space dimensions to each user in the TIM component, and guarantees that theproduct of the two is an achievable number of signal dimensions available to eachuser in the original network.

10:10

On the Interference Channel with Common Messages and theRole of Rate-SharingStefano Rini (Stanford, USA); Andrea Goldsmith (Stanford University,USA)

The capacity region of the interference channel with common messages is studied.This channel model is a variation of the classical two user interference channelmodified such that each encoder has both a private message as well as a com-mon message to be decoded at both receivers. Achievable rates for this channelmodel can be characterized by the classical Han-Kobayashi achievable rate regionfor the interference channel in which, at each encoder, the codeword embeddingthe private message is superimposed over the codeword for the common message.We show that the achievable rates for this region can be improved upon by rate-sharing, which consist of transmitting part of the private message into the common

codeword. This improved region is shown to approach the capacity for a class of in-jective semi-deterministic channels. Moreover, we show that the Fourier-Motzkinelimination of the Han-Kobayashi with rate-sharing contains less rate bounds thanthe one without rate-sharing. This approach provides an alternative proof of thesimplification of the Han-Kobayashi originally shown by Chong et al. This resultis particularly interesting as it shows that simplifications in the spirit of Chong et al.for general channels can be performed through rate-sharing and Fourier-Motzkinelimination. This approach can be easily implemented algorithmically and is rele-vant in the context of the automatic derivation of achievable rate regions.

10:30

The Stability Region of the Two-User Interference ChannelNikolaos Pappas (Supelec, France); Marios Kountouris (Supelec,France); Anthony Ephremides (University of Maryland at College Park,USA)

The stable throughput region of the two-user interference channel is investigatedhere. First, the stability region for the general case is characterized. Second, westudy the cases where the receivers treat interference as noise or perform suc-cessive interference cancelation. Finally, we provide conditions for the convex-ity/concavity of the stability region and for which a certain interference manage-ment strategy leads to broader stability region.

10:50

On the Reliable Transmission of Correlated Sources OverTwo-Relay NetworkMohammad Nasiraee (K. N. Toosi University of Technology, Iran); Ba-hareh Akhbari (K. N. Toosi University of Technology, Iran); MahmoudAhmadian (K. N. Toosi University of Technology, Iran); Mohammad RezaAref (Sharif University of Tech., Iran)

In this paper, we investigate reliable transmission of three correlated discrete mem-oryless sources over a two-relay network. In our considered model, one of thesources is available at the sender whereas, the other two sources are known to thefirst and the second relay. We present both joint and separate source-channel cod-ing schemes, and derive the corresponding sets of sufficient conditions for reliablesources transmission. The manner of cooperation in both schemes is Decode-and-Forward strategy. In the joint approach, we generalize the correlation preservingmapping technique to our model using nested backward decoding. Our proposedseparate approach is based on Slepian-Wolf source coding and irregular encod-ing/successive decoding strategy. Furthermore, we obtain necessary conditions forreliable sources transmission over the network. Our results can be reduced to theseveral known results in the literature.

Thursday, 11:30–12:50


ThC1: Source Coding

Chair: Ram Zamir (Tel Aviv University, Israel)

11:30

Nonasymptotic Noisy Source CodingVictoria Kostina (Princeton University, USA); Sergio Verdu (PrincetonUniversity, USA)

This paper shows new general nonasymptotic achievability and converse boundsand performs their dispersion analysis for the lossy compression problem in whichthe compressor observes the source through a noisy channel. While this problem isasymptotically equivalent to a noiseless lossy source coding problem with a mod-ified distortion function, nonasymptotically there is a difference in how fast theirminimum achievable coding rates approach the rate-distortion function, providingyet another example where at finite blocklengths one must put aside traditionalasymptotic thinking.

11:50

The Heegard-Berger Problem with Common Receiver Recon-structions

34


Badri N. Vellambi (University of South Australia, Australia); Roy Timo(University of South Australia, Australia)

The variant of the Heegard-Berger problem where the receivers require identicalreconstructions is studied, and the rate-distortion function for the following threesettings is derived: (a) the encoder is also required to generate the reconstructionscommon to the receivers; (b) the side information at the receivers are physicallydegraded; and (c) the side information at the receivers are stochastically degraded,and the source satisfies a particular full-support condition. The characterizationsindicate that the receiver side information can be fully exploited to perform Wyner-Ziv-style binning. However, the reconstruction functions can depend only on therandomness common to the receivers in the Gacs-Korner sense.

12:10

Lattice Quantization Noise RevisitedCong Ling (Imperial College London, United Kingdom); Lu Gan (BrunelUniversity, United Kingdom)

Dithered quantization is widely used in signal processing and coding due to itsmany desirable properties in theory and in practice. In this paper, we show thatdither is unnecessary for Gaussian sources when the flatness factor of the quanti-zation lattice is small. This means that the quantization noise behaves much likethat in dithered quantization. In particular, it tends to be uniformly distributed overany fundamental region of the lattice and be uncorrelated with the signal; further,for optimum lattice quantizers, it approaches the rate-distortion bound of Gaussiansources with minimum mean-square error (MMSE) estimation.

12:30

The Likelihood Encoder for Source CodingPaul Cuff (Princeton University, USA); Eva Chen Song (Princeton Uni-versity, USA)

The likelihood encoder with a random codebook is demonstrated as an effectivetool for source coding. Coupled with a soft covering lemma (associated with chan-nel resolvability), likelihood encoders yield simple achievability proofs for knownresults, such as rate-distortion theory. They also produce a tractable analysis forsecure rate-distortion theory and strong coordination.


ThC2: Information and Estimation

Chair: Laszlo Gyorfi (Budapest University of Technology and Eco-nomics, Hungary)

11:30

Non-Parametric Prediction of the Mid-Price Dynamics in aLimit Order BookDeepan Palguna (Purdue University, USA); Ilya Pollak (Purdue Univer-sity, USA)

Many securities markets are organized as double auctions where each incominglimit order—i.e., an order to buy or sell at a specific price—is stored in a datastructure called the limit order book. A trade happens whenever a marketable orderarrives. This order flow is visible to every market participant in real time. Wepropose a novel non-parametric approach to short-term forecasting of the mid-pricechange in a limit order book (i.e., of the change in the average of the best offer andthe best bid prices). We construct a state characterizing the order book at each timeinstant and compute a feature vector for each value of the state. The features getupdated during the course of a trading day, as new order flow information arrives.Our prediction rules at every time instant during the trading day are based on thefeature vector of the state observed at that time instant. The distinction of ourapproach from the previous ones is that it does not impose a restrictive parametricmodel. Implicit assumptions of our method are very mild. Initial experiments withreal order book data from NYSE suggest that our algorithms show promise. Weillustrate their usage in a practical application of executing a large trade.

11:50

One-shot bounds for various information theoretic problemsusing smooth min and max Renyi divergences

Naqueeb Warsi (Tata Institute of Fundamental Research, India)

One-shot analogues for various information theory results known in the asymptoticcase are proven using smooth min and max Renyi divergences. In particular, weprove that smooth min Renyi divergence can be used to prove one-shot analogue ofthe Stein’s lemma. Using smooth min Renyi divergence we prove a special case ofpacking lemma in the one-shot setting. Furthermore, we prove a one-shot analogueof covering lemma using smooth max Renyi divergence. We also propose one-shot achievable rate for source coding under maximum distortion criterion. Thisachievable rate is quantified in terms of smooth max Renyi divergence.

12:10

Signals that can be easily time-frequency synchronized fromtheir ambiguity functionTohru Kohda (Kyushu University, Japan); Yutaka Jitsumatsu (KyushuUniversity, Japan); Kazuyuki Aihara (University of Tokyo, Japan)

Delay and Doppler determination of a time-delayed and frequency-shifted signal isone of fundamental problems in communication. The two-parameter estimation isreduced to two one-parameter estimation problems in time- and frequency-domainsignals. Motivated by Gabor’s communication theory, we proceed further paral-lelism between time- and frequency-domain signals and solve the two problemsindividually and cooperatively. Simulation results without prescribed informationare reported.

12:30

Novel Tight Classification Error Bounds under MismatchConditions based on f -DivergenceRalf Schluter (RWTH Aachen University, Germany); Markus Nussbaum-Thom (RWTH Aachen University, Germany); Eugen Beck (RWTH AachenUniversity, Germany); Tamer Alkhouli (RWTH Aachen University, Ger-many); Hermann Ney (RWTH Aachen, Germany)

By default, statistical classification/multiple hypothesis testing is faced with themodel mismatch introduced by replacing the true distributions in Bayes decisionrule by model distributions estimated on training samples. Although a large num-ber of statistical measures exist w.r.t. to the mismatch introduced, these worksrarely relate to the mismatch in accuracy, i.e. the difference between model er-ror and Bayes error. In this work, the accuracy mismatch between the ideal Bayesdecision rule/Bayes test and a mismatched decision rule in statistical classifica-tion/multiple hypothesis testing is investigated explicitly. A proof of a novel gen-eralized tight statistical bound on the accuracy mismatch is presented. This resultis compared to existing statistical bounds related to the total variational distancethat can be extended to bounds of the accuracy mismatch. The analytic results aresupported by distribution simulations.


ThC3: Interference Networks

Chair: Syed Ali Jafar (University of California Irvine, USA)

11:30

Degrees of Freedom of the Rank-Deficient MIMO X channelAdrian Agustin (Universitat Politecnica de Catalunya (UPC), Spain);Josep Vidal (Universitat Politecnica de Catalunya, Spain)

The multiple-input multiple-output (MIMO) X channel with rank-deficient chan-nels is considered. We characterize the total degrees of freedom (DoF) by definingan outer bound which is attained by the proposed precoding scheme. The total DoFare described in closed-form when all transmitters and receivers have M and Nantennas, respectively. For an arbitrary number of antennas, the total DoF are ob-tained as solution of a linear programming problem. Results elucidate that, unlikepoint-to-point MIMO systems where DoF increase with the rank of the channel,the total DoF in a multipoint-to-multipoint MIMO system can increase up to a cer-tain point if channels are rank-deficient. Hence, we corroborate the relevance ofknowing the rank-deficiency in the MIMO X channel, a property already observedin the MIMO Interference channel.

11:50

Interference Neutralization using Lattice Codes

35


Shahab Ghasemi-Goojani (Sharif University of Tech., Iran); HamidBehroozi (Sharif University of Tech., Iran)

Deterministic approach of [1] models the interaction between the bits that arereceived at the same signal level by the modulo 2 sum of the bits where thecarry-overs that would happen with real addition are ignored. By this model ina multi-user setting, the receiver can distinguish most significant bits (MSBs) ofthe stronger user without any noise. A faithful implementation of the determinis-tic model requires one to “neutralize interference” from previous carry over digits.This paper proposes a new implementation of “interference neutralization” [2] us-ing structured lattice codes. We first present our implementation strategy and then,as an application, apply this strategy to a symmetric half-duplex Gaussian butterflynetwork. In this network, two transmitters communicate their information to twodestinations via a half-duplex relay. For duplexing factor 0.5 and under certainconditions, we show that our proposed scheme based on superposition of nestedlattice codes can achieve the capacity region in high SNR. Also, regardless of allchannel parameters, the gap between its achievable rate region and the outer boundis at most 0.293 bits/sec/Hz.

12:10

Cognitive Cooperative Communications on the Multiple Ac-cess ChannelJonathan Shimonovich (Technion - Israel Institute of Technology, Israel);Anelia Somekh-Baruch (Bar-Ilan University, Israel); Shlomo (Shitz)Shamai (The Technion, Israel)

In this paper, we investigate a three-user cognitive communication network wherea primary two-user Multiple Access Channel (MAC) suffers interference from asecondary point-to-point (P2P) channel, sharing the same medium. While the P2Pchannel transmitter —–transmitter 3–— causes an interference at the primary MACreceiver, we assume that the primary channel transmitters —–transmitters 1 and 2–— do not cause any interference at the P2P receiver. It is assumed that one ofthe MAC transmitters has cognitive capabilities and cribs causally from the otherMAC transmitter. Furthermore, we assume that the cognitive transmitter knowsthe message of transmitter 3 in a noncausal manner, thus introducing the three-user Multiple Access Cognitive Z-Interference Channel (MA- CZIC). We obtaininner and outer bounds on the capacity region of the three-user MA-CZIC for bothstrictly and nonstrictly causal cribbing cognitive encoders.

12:30

State-Dependent Gaussian Z-Channel with Mismatched Side-Information and InterferenceRuchen Duan (Syracuse University, USA); Yingbin Liang (Syracuse Uni-versity, USA); Ashish Khisti (University of Toronto, Canada); Shlomo(Shitz) Shamai (The Technion, Israel)

A state-dependent Gaussian Z-interference channel model is investigated in theregime of high state power, in which transmitters 1 and 2 communicate with re-ceivers 1 and 2, and only receiver 2 is interfered by transmitter 1’s signal and a ran-dom state sequence. The state sequence is known noncausally only to transmitter1, not to the corresponding transmitter 2. A layered coding scheme is designed fortransmitter 1 to help interference cancelation at receiver 2 (using a cognitive dirtypaper coding) and to transmit its own message to receiver 1. Inner and outer boundsare derived, and are further analyzed to characterize the boundary of the capacityregion either fully or partially for all Gaussian channel parameters. Our resultsimply that the capacity region of such a channel with mismatched transmitter-sidestate cognition and receiver-side state interference is strictly less than that of thecorresponding channel without state, which is in contrast to Costa type of dirtychannels, for which dirty paper coding achieves the capacity of the correspondingchannels without state.

Thursday’s Plenaritas, 14:00–15:00

Thursday, 14:00–15:00 Salon de Actos

Chair: Daniel J. Costello, Jr. (University of Notre Dame, USA)

14:00

Detection of Correlations and Random Geometric Graphs

Gabor Lugosi (Universitat Pompeu Fabra, Spain)

In this talk we explore the the limits of testing whether a correlation matrix of amultivariate normal population is the identity matrix. We consider on sparse classesof alternatives where only a few entries are nonzero and, in fact, positive. We derivea general lower bound applicable to various classes and study the performance ofsome near-optimal tests. The detection problem naturally leads to the study of theclique number of high-dimensional random geometric graphs. Interestingly, lowerbounds for the risk of the likelihood ratio test lead to nontrivial bounds for theclique number.(Joint work with Sebastien Bubeck (Princeton) and Ery Arias-Castro (UCSD))

Bio: Gabor Lugosi graduated in electrical engineering at the Technical Univer-sity of Budapest in 1987, and received his Ph. D. from the Hungarian Academyof Sciences in 1991. Since 1996, he has been at the Department of Economics,Pompeu Fabra University. In 2006 he became an ICREA research professor. Hisresearch interest includes learning theory, nonparametric statistics, inequalities inprobability, random structures, and information theory.

14:30

Coding for Chip-to-Chip CommunicationAmin Shokrollahi (Kandou Bus and EPFL, Switzerland)

Modern electronic devices consist of a multitude of IC components: the proces-sor, the memory, the Modem (in wireless devices), the graphics processor, are onlysome examples of components scattered throughout a device. The increase of thevolume of digital data that needs to be accessed and processed by such devicescalls for ever faster communication between these IC’s. Faster communication,however, often translates to higher susceptibility to various types of noise, and in-evitably to a higher power consumption in order to combat noise. This increasein the power consumption is, for the most part, far from linear, and cannot be eas-ily compensated for by Moore’s Law. In this talk I will give a short overview ofproblems encountered in chip-to-chip communication, and will advocate the use ofnovel coding techniques to solve those problems. I will also briefly talk about Kan-dou Bus, and some of the approaches the company is taking to design, implement,and market such solutions.

Bio: Amin Shokrollahi finished his PhD at the University of Bonn in 1991. In 1995he joined the International Computer Science Institute in Berkeley and in 1998he joined the Bell Laboratories. In 2000 he became the Chief Scientist of DigitalFountain, a position he held until Digital Fountain’s acquisition by Qualcomm in2009. In 2003 he joined EPFL where is a full professor of Mathematics and Com-puter Science. In 2011 he founded the company Kandou Bus which uses novelcoding approaches for the design of fast and energy efficient chip-to-chip links.Currently, he is the CEO of that company. Amin’s research interests are varied andcover coding theory, discrete mathematics, algorithm design, theoretical computerscience, signal processing, networking, computational algebra, algebraic complex-ity theory, and most recently, electronics. He has more than 120 publications, andmore than 70 pending and granted patent applications in these areas. An IEEEFellow, Amin’s honors include the IEEE Information Theory Society InformationTheory Paper Award (2002), the joint IT/ComSoc Best Paper Award (2008), theIEEE Eric E. Sumner Communications Theory Award (2007), the IEEE HammingMedal (2012), and the Advanced Research Grant of the European Research Coun-cil.

Thursday, 15:10–16:30


ThE1: Codes for Storage

Chair: Hendrik C. Ferreira (University of Johannesburg, South Africa)

15:10

Exact-Regenerating Codes between MBR and MSR PointsToni Ernvall (University of Turku, Finland)

In this paper we study distributed storage systems with exact repair. We give a con-struction for regenerating codes between the minimum storage regenerating (MSR)and the minimum bandwidth regenerating (MBR) points and show that in the case

36


that the parameters n, k, and d are close to each other our constructions are closeto optimal when comparing to the known capacity when only functional repair isrequired. We do this by showing that when the distances of the parameters n, k,and d are fixed but the actual values approach to infinity, the fraction of the per-formance of our codes with exact repair and the known capacity of codes withfunctional repair approaches to one.

15:30

Prefixless q-ary Balanced Codes with ECCTheo G. Swart (University of Johannesburg, South Africa); Kees A.Schouhamer Immink (Turing Machines Inc., The Netherlands)

We present a Knuth-like method for balancing q-ary codewords, which is char-acterized by the absence of a prefix that carries the information of the balancingindex. Look-up tables for coding and decoding the prefix are avoided. We alsoshow that this method can be extended to include error correction of single channelerrors.

15:50

Constrained Rank Modulation SchemesFrederic Sala (University of California, Los Angeles, USA); Lara Dolecek(UCLA, USA)

Rank modulation schemes for non-volatile memories (NVMs) represent informa-tion by the relative rankings of cell charge levels. This approach has several bene-fits; in particular, the scheme resolves the “write-asymmetry” limitation that NVMssuffer from. However, cell writing is still affected by a common NVM problem:inter-cell coupling, which can result in inadvertently increasing the charge levelof neighboring cells. This is a potential source of error in the rank modulationscheme.In this paper, we explore the idea of constrained coding over permutations. Theseconstraints minimize the impact of inter-cell coupling while still allowing the useof the rank modulation scheme. We study various constraints and their resultingrates, capacities, and other properties, and introduce an explicit constrained rankmodulation code construction.

16:10

Maximum Likelihood Associative MemoriesVincent Gripon (Telecom Bretagne, France); Michael Rabbat (McGillUniversity, Canada)

Associative memories are structures that store data in such a way that it can laterbe retrieved given only a part of its content -— a sort-of error/erasure-resilienceproperty. They are used in applications ranging from caches and memory manage-ment in CPUs to database engines. In this work we study associative memoriesbuilt on the maximum likelihood principle. We derive minimum residual errorrates when the data stored comes from a uniform binary source. Second, we deter-mine the minimum amount of memory required to store the same data. Finally, webound the computational complexity for message retrieval. We then compare thesebounds with two existing associative memory architectures: the celebrated Hop-field neural networks and a neural network architecture introduced more recentlyby Gripon and Berrou.


ThE2: Big Data: Statistics and Information Theory in High Dimen-sions and Undersampled Regimes(Organized by Prasad Santhanam)

Chair: Prasad Santhanam (University of Hawaii at Manoa, USA)

15:10

Informational Confidence Bounds for Self-Normalized Aver-ages and ApplicationsAurelien Garivier (Universite Paul Sabatier Toulouse, France)

We present deviation bounds for self-normalized averages and applications to es-timation with a random number of observations. The results rely on a peelingargument in exponential martingale techniques that represents an alternative to themethod of mixture. The motivating examples of bandit problems and context tree

estimation are detailed.

15:30

MCUIUC — A New Framework for Metagenomic Read Com-pressionJonathan Ligo (UIUC, USA); Minji Kim (University of Illinois atUrbana-Champaign, USA); Amin Emad (University of Illinois at Urbana-Champaign, USA); Olgica Milenkovic (UIUC, USA); Venugopal Veer-avalli (University of Illinois at Urbana-Champaign, USA)

Metagenomics is an emerging field of molecular biology concerned with analyz-ing the genomes of environmental samples comprising many different diverse or-ganisms. Given the nature of metagenomic data, one usually has to sequence thegenomic material of all organisms in a batch, leading to a mix of reads com-ing from different DNA sequences. In deep high-throughput sequencing experi-ments, the volume of the raw reads is extremely high, frequently exceeding 600Gb. With an ever increasing demand for storing such reads for future studies, theissue of efficient metagenomic compression becomes of paramount importance.We present the first known approach to metagenome read compression, termedMCUIUC (Metagenomic Compression at UIUC). The gist of the proposed algo-rithm is to perform classification of reads based on unique organism identifiers,followed by reference-based alignment of reads for individually identified organ-isms, and metagenomic assembly of unclassified reads. Once assembly and clas-sification are completed, lossless reference based compression is performed viapositional encoding. We evaluate the performance of the algorithm on moderatesized synthetic metagenomic samples involving 15 randomly selected organismsand describe future directions for improving the proposed compression method.

15:50

Sparse Regression Codes: Recent Results and Future Direc-tionsRamji Venkataramanan (University of Cambridge, United Kingdom);Sekhar Tatikonda (Yale University, USA)

Sparse Superposition or Sparse Regression codes were recently introduced by Bar-ron and Joseph for communication over the AWGN channel. The code is defined interms of a design matrix; codewords are linear combinations of subsets of columnsof the matrix. These codes achieve the AWGN channel capacity with computa-tionally feasible decoding. We have shown that they also achieve the optimal rate-distortion function for Gaussian sources. Further, the sparse regression codebookhas a partitioned structure that facilitates random binning and superposition. In thispaper, we review existing results concerning Sparse Regression codes and discussdirections for future research.

16:10

Estimation of transition and stationary probabilities of slowmixing Markov processesNarayana Prasad Santhanam (University of Hawaii at Manoa, USA);Meysam Asadi (University of Hawaii at Manoa, USA); Ramezan ParaviTorghabeh (University of Hawaii at Manoa, USA)

We observe a length-n sample generated by an unknown, stationary ergodicMarkov process (model) over a finite alphabet A. Motivated by applications inwhat are known as backplane channels, given any string w of symbols from A wewant estimates of the conditional probability distribution of symbols following w(model parameters), as well as the stationary probability of w.Two distinct problems that complicate estimation in this setting are (i) long mem-ory, and (ii) slow mixing which could happen even with only one bit of memory.Any consistent estimator can only converge pointwise over the class of all ergodicMarkov models. Namely, given any estimator and any sample size n, the under-lying model could be such that the estimator performs poorly on a sample of sizen with high probability. But can we look at a length-n sample and identify if anestimate is likely to be accurate?Since the memory is unknown a-priori, a natural approach is to estimate a poten-tially coarser model with memory kn = O(logn). As n grows, estimates getrefined and this approach is consistent with the above scaling of kn also knownto be essentially optimal. But while effective asymptotically, the situation is quitedifferent when we want the best answers possible with a length-n sample, ratherthan just consistency. Combining results in universal compression with Aldous’coupling arguments, we obtain sufficient conditions on the length-n sample (evenfor slow mixing models) to identify when naive (i) estimates of the model param-

37


eters and (ii) estimates related to the stationary probabilities are accurate; and alsobound the deviations of the naive estimates from true values.


ThE3: Source Coding in Networks

Chair: J. Nicholas Laneman (University of Notre Dame, USA)

15:10

On Existence of Optimal Linear Encoders over Non-fieldRings for Data Compression with Application to ComputingSheng Huang (KTH Royal Institute of Technology, Sweden); MikaelSkoglund (KTH Royal Institute of Technology, Sweden)

This note proves that, for any finite set of correlated discrete i.i.d. sources, therealways exists a sequence of linear encoders over some finite non-field rings whichachieves the data compression limit, the Slepian–Wolf region. Based on this, weaddress a variation of the data compression problem which considers recoveringsome discrete function of the data. It is demonstrated that linear encoder over non-field ring strictly outperforms its field counterpart for encoding some function interms of achieving strictly larger achievable region with strictly smaller alphabetsize.

15:30

Sampling versus Random Binning for Multiple Descriptions ofa Bandlimited SourceAdam Mashiach (Tel-Aviv University, Israel); Jan Østergaard (AalborgUniversity, Denmark); Ram Zamir (Tel Aviv University, Israel)

Random binning is an efficient, yet complex, coding technique for the symmet-ric L-description source coding problem. We propose an alternative approach,that uses the quantized samples of a bandlimited source as “descriptions”. By theNyquist condition, the source can be reconstructed if enough samples are received.We examine a coding scheme that combines sampling and noise-shaped quantiza-tion for a scenario in which only K < L descriptions or all L descriptions arereceived. Some of the received K-sets of descriptions correspond to uniform sam-pling while others to non-uniform sampling. This scheme achieves the optimumrate-distortion performance for uniform-sampling K-sets, but suffers noise ampli-fication for nonuniform-sampling K-sets. We then show that by increasing thesampling rate and adding a random-binning stage, the optimal operation point isachieved for any K-set.

15:50

Capacity Region of Multi-Resolution Streaming in Peer-to-Peer NetworksBatuhan Karagoz (Middle East Technical University, Turkey); SemihYavuz (Bilkent University, Turkey); Tracey Ho (California Institute ofTechnology, USA); Michelle Effros (California Institute of Technology,USA)

We consider multi-resolution streaming in fully-connected peer-to-peer networks,where transmission rates are constrained by arbitrarily specified upload capacitiesof the source and peers. We fully characterize the capacity region of rate vectorsachievable with arbitrary coding, where an achievable rate vector describes a vectorof throughputs of the different resolutions that can be supported by the network. Wethen prove that all rate vectors in the capacity region can be achieved using purerouting strategies. This shows that coding has no capacity advantage over routingin this scenario.

16:10

A Deterministic Annealing Approach to Optimization of Zero-delay Source-Channel CodesMustafa S. Mehmetoglu (University of California, Santa Barbara, USA);Emrah Akyol (UCSB, USA); Kenneth Rose (University of California,Santa Barbara, USA)

This paper studies optimization of zero-delay source-channel codes, and specifi-cally the problem of obtaining globally optimal transformations that map between

the source space and the channel space, under a given transmission power con-straint and for the mean square error distortion. Particularly, we focus on the set-ting where the decoder has access to side information, whose cost surface is knownto be riddled with local minima. Prior work derived the necessary conditions foroptimality of the encoder and decoder mappings, along with a greedy optimizationalgorithm that imposes these conditions iteratively, in conjunction with the heuris-tic “noisy channel relaxation” method to mitigate poor local minima. While noisychannel relaxation is arguably effective in simple settings, it fails to provide accu-rate global optimization results in more complicated settings including the decoderwith side information as considered in this paper. We propose a global optimizationalgorithm based on the ideas of “deterministic annealing”– a non-convex optimiza-tion method, derived from information theoretic principles with analogies to statis-tical physics, and successfully employed in several problems including clustering,vector quantization and regression. We present comparative numerical results thatshow strict superiority of the proposed algorithm over greedy optimization methodsas well as over the noisy channel relaxation.

Thursday, 16:50–18:10


ThF1: Source and Channel Coding

Chair: Ashish Khisti (University of Toronto, Canada)

16:50

The Least Degraded and the Least Upgraded Channel withrespect to a Channel FamilyWei Liu (EPFL, Switzerland); S. Hamed Hassani (EPFL, Switzerland);Rudiger Urbanke (EPFL, Switzerland)

Given a family of binary-input memoryless output-symmetric (BMS) channelshaving a fixed capacity, we derive the BMS channel having the highest (resp. low-est) capacity among all channels that are degraded (resp. upgraded) with respect tothe whole family. We give an explicit characterization of this channel as well as anexplicit formula for the capacity of this channel.

17:10

On 2-D Non-Adjacent-Error Channel ModelsShivkumar Manickam (Indian Institute of Science, India); Navin Kashyap(Indian Institute of Science, India)

In this work, we consider two-dimensional (2-D) binary channels in which the 2-Derror patterns are constrained so that errors cannot occur in adjacent horizontal orvertical positions. We consider probabilistic and combinatorial models for suchchannels. A probabilistic model is obtained from a 2-D random field defined byRoth, Siegel and Wolf (2001). Based on the conjectured ergodicity of this ran-dom field, we obtain an expression for the capacity of the 2-D non-adjacent-errorschannel. We also derive an upper bound for the asymptotic coding rate in the com-binatorial model.

17:30

On the Mutual Information between Random Variables inNetworksXiaoli Xu (Nanyang Technological University, Singapore); SatyajitThakor (Institute of Network Coding, The Chinese University of HongKong, Hong Kong); Yong Liang Guan (Nanyang Technological Univer-sity, Singapore)

This paper presents a lower bound on the mutual information between any twosets of source/edge random variables in a general multi-source multi-sink network.This bound is useful to derive a new class of better information-theoretic upperbounds on the network coding capacity given existing edge-cut based bounds. Arefined functional dependence bound is characterized from the functional depen-dence bound using the lower bound. It is demonstrated that the refined versionsof the existing edge-cut based outer bounds obtained using the mutual informationlower bound are stronger.

17:50

38


A Connection between Good Rate-distortion Codes and Back-ward DMCsCurt Schieler (Princeton University, USA); Paul Cuff (Princeton Univer-sity, USA)

Let Xn ∈ Xn be a sequence drawn from a discrete memoryless source, andlet Y n ∈ Yn be the corresponding reconstruction sequence that is output by agood rate-distortion code. This paper establishes a property of the joint distri-bution of (Xn, Y n). It is shown that for D > 0, the input-output statistics ofa R(D)-achieving rate-distortion code converge (in normalized relative entropy)to the output-input statistics of a discrete memoryless channel (dmc). The dmc is“backward” in that it is a channel from the reconstruction spaceYn to source spaceXn. It is also shown that the property does not necessarily hold when normalizedrelative entropy is replaced by variational distance.


ThF2: Information Theoretic Problems in Computer Science(Organized by Venkat Anantharam)

Chair: Venkat Anantharam (University of California at Berkeley, USA)

16:50

Bypassing Correlation Decay for Matchings with an Applica-tion to XORSATMarc Lelarge (INRIA and ENS, France)

Many combinatorial optimization problems on sparse graphs do not exhibit the cor-relation decay property. In such cases, the cavity method remains a sophisticatedheuristic with no rigorous proof. In this paper, we consider the maximum matchingproblem which is one of the simplest such example. We show that monotonicityproperties of the problem allows us to define solutions for the cavity equations.More importantly, we are able to identify the ‘right’ solution of these equationsand then to compute the asymptotics for the size of a maximum matching. Theresults for finite graphs are self-contained. We give references to recent extensionsmaking use of the notion of local weak convergence for graphs and the theory ofunimodular networks.As an application, we consider the random XORSAT problem which accordingto the physics literature has a ‘one-step replica symmetry breaking’ (1RSB) glassphase. We derive new bounds on the satisfiability threshold valid for general graphs(and conjectured to be tight).

17:10

Lovasz ϑ, SVMs and ApplicationsVinay Jethava (Chalmers University of Technology, Sweden); Jacob Sz-najdman (Chalmers University, India); Chiranjib Bhattacharya (IISc, In-dia); Devdatt Dubhashi (Chalmers University of Technology, Sweden)

The Lovasz theta function was introduced to solve the problem of computing theShannon capacity of the pentagon. It has subsequently become a fundamental toolin information theory, graph theory and combinatorial optimization. However com-puting it requires solving a SDP, making it hard to scale to large graphs. Recentlywe discovered a connection with the one class support vector machine (SVM) for-mulation, which yields an approximation to the Lovasz function that can scale tolarge graphs. Here we review this connection and give a number of applications tofundamental graph theory and data mining problems such as coloring, max–k–cutand large sparse/dense subgraphs.

17:30

Local Privacy, Quantitative Data Processing Inequalities, andStatistical Minimax RatesJohn Duchi (UC Berkeley, USA); Michael Jordan (UC Berkeley, USA);Martin J. Wainwright (University of California, Berkeley, USA)

Working under local differential privacy—a model of privacy in which data re-mains private even from the statistician or learner—we study the tradeoff betweenprivacy guarantees and the utility of the resulting statistical estimators. We showthat local privacy acts as a type of contraction for information theoretic quantitiesincluding mutual information and Kullback-Leibler divergence, which thus influ-ence estimation rates as a function of the amount of privacy preserved. Our bounds

can be viewed as quantitative data-processing inequalities. When combined withminimax techniques such as Le Cam’s and Fano’s methods, these inequalities al-low for a precise characterization of statistical rates under local privacy constraints.In this paper, we provide a complete treatment of two canonical problem families:mean estimation in location family models and estimation of multinomial proba-bilities. For these families, we provide lower and upper bounds that match up toconstant factors.

17:50

On the Information Complexity of Cascaded Norms withSmall DomainsT. S. Jayram (IBM Almaden Research Center, USA)

We consider the problem of estimating cascaded norms in a data stream, a well-studied generalization of the classical norm estimation problem, where the data isaggregated in a cascaded fashion along multiple attributes. We show that whenthe number of attributes for each item is at most d, then estimating the cascadednorm Lk L1 requires space Ω(dn1−2/k) for every d = O(n1/k). This resultinterpolates between the tight lower bounds known previously for the two extremesof d = 1 and d = Θ(n1/k) [1]. The proof of this result uses the informationcomplexity paradigm that has proved successful in obtaining tight lower boundsfor several well-known problems. We use the above data stream problem as amotivation to sketch some of the key ideas of this paradigm. In particular, we givea unified and a more general view of the key negative-type inequalities satisfied bythe transcript distributions of communication protocols.


ThF3: Multicell and Broadcast

Chair: Giuseppe Caire (University of Southern California, USA)

16:50

Uplink Multi-Cell Processing: Approximate Sum Capacityunder a Sum Backhaul ConstrainYuhan Zhou (University of Toronto, Canada); Wei Yu (University ofToronto, Canada); Dimitris Toumpakaris (University of Patras, Greece)

This paper investigates an uplink multi-cell processing (MCP) model where thecell sites are linked to a central processor (CP) via noiseless backhaul links withlimited capacity. A simple compress-and-forward scheme is employed, where thebase-stations (BSs) quantize the received signals and send the quantized signals tothe CP using distributed Wyner-Ziv compression. The CP decodes the quantizationcodewords first, then decodes the user messages as if the users and the CP form avirtual multiple-access channel. This paper formulates the problem of maximizingthe overall sum rate under a sum backhaul constraint for such a setting. It is shownthat setting the quantization noise levels to be uniform across the BSs maximizesthe achievable sum rate under high signal-to-noise ratio (SNR). Further, for generalSNR a low-complexity fixed-point iteration algorithm is proposed to optimize thequantization noise levels. This paper further shows that with uniform quantizationnoise levels, the compress-and-forward scheme with Wyner-Ziv compression al-ready achieves a sum rate that is within a constant gap to the sum capacity of theuplink MCP model. The gap depends linearly on the number of BSs in the networkbut is independent of the SNR and the channel matrix.

17:10

Multi-Cell Cooperation with Random User Locations underArbitrary SignalingMaksym A. Girnyk (KTH Royal Institute of Technology, Sweden); MikkoVehkapera (Aalto University, Finland); Lars K. Rasmussen (KTH RoyalInstitute of Technology, Sweden)

Base station cooperation in cellular networks has been recently recognized as a keytechnology for mitigating interference, providing thus significant improvements inthe system performance. In this paper, we consider a simple scenario consisting oftwo one-dimensional cells, where the base stations have fixed locations, while theuser terminals are randomly distributed on a line. Exploiting the replica methodfrom statistical physics, we derive the ergodic sum-rate under arbitrary signalingfor both cooperative and non-cooperative scenarios, when the system size growslarge. The obtained results are analytically tractable and can be used to optimizethe system parameters in a simple manner. The numerical examples show that the

39


analysis provides good approximations for finite-sized systems.

17:30

Any Positive Feedback Rate Increases the Capacity of StrictlyLess-Noisy Broadcast ChannelsYoulong Wu (Telecom ParisTech, France); Michele Wigger (TelecomParisTech, France)

We propose two coding schemes for discrete memoryless broadcast channels (DM-BCs) with rate-limited feedback from only one receiver. For any positive feedbackrate and for the class of strictly less-noisy DMBCs, our schemes strictly improveover the no-feedback capacity region.

17:50

On the Role of Interference Decoding in Compound BroadcastChannelsMeryem Benammar (Supelec, France); Pablo Piantanida (Supelec,France)

This work investigates the general three-message compound broadcast channel(BC) where an encoder wishes to communicate a common and two private mes-sages to two groups of users. We focus on the study of the largest achievableregion when the encoder is constrained to use a single description per message thatwe refer to as “simplified random codes”. In this setting, we investigate the differ-ent roles that decoders can play at the destinations. Surprisingly enough, we showthat simultaneous decoding of both the intended as well as the interference mes-sage at the destinations can strictly enlarge –in opposition with the standard BC–the rate region with respect to the worst-case (over all channels) of Marton’s innerbound.

Conference Banquet

Thursday, 20:30–23:00 Restaurante Abades (C/ Betis, 69)

Friday

Friday’s Plenary, 08:40–09:40

Friday, 08:40–09:40 Salon de Actos

Chair: Gerhard Kramer (Technische Universitat Munchen, Germany)

Strategic Computation in NetworksMichael Kearns (University of Pennsylvania, USA)

What do the theory of computation, economics and related fields have to say aboutthe emerging phenomena of crowdsourcing and social computing? Most successfulapplications of crowdsourcing to date have been on problems we might consider“embarrassingly parallelizable” from a computational perspective. But the powerof the social computation approach is already evident, and the road cleared forapplying it to more challenging problems.In part towards this goal, for a number of years we have been conducting con-trolled human-subject experiments in distributed social computation in networkswith only limited and local communication. These experiments cast a number oftraditional computational problems — including graph coloring, consensus, inde-pendent set, market equilibria, biased voting and network formation — as gamesof strategic interaction in which subjects have financial incentives to collectively“compute” global solutions. I will overview and summarize the many behavioralfindings from this line of experimentation, and draw broad comparisons to some ofthe predictions made by the theory of computation and microeconomics.

Bio: Michael Kearns is Professor and National Center Chair in the Computer andInformation Science department at the University of Pennsylvania, where he holdssecondary appointments in the Statistics and Operations and Information Man-agement departments of the Wharton School. His primary research interests are

in machine learning, algorithmic game theory, social networks and related topics.He is the Founding Director of Penn’s Networked and Social Systems Engineer-ing (NETS) undergraduate program (www.nets.upenn.edu), and until 2006 was di-rector of Penn’s Institute for Research in Cognitive Science. Prior to joining thePenn faculty in 2002, Kearns spent a decade in basic computer science research atAT&T/Bell Labs, where he headed the Artificial Intelligence research department.He has received a variety of awards for his research, including an ACM Distin-guished Dissertation Award, the Henry and Bryna David Award from the NationalAcademy of Sciences, and election as a Fellow of the American Academy of Artsand Sciences.

Friday, 09:50–11:10

Friday, 09:50–11:10 Room 1 – Sala Juan Larraneta

FrB1: Information Function Computation

Chair: Maxim Raginsky (University of Illinois at Urbana-Champaign,USA)

09:50

Distributed Function Computation Over a Tree NetworkMilad Sefidgaran (Telecom ParisTech, France); Aslan Tchamkerten (Tele-com ParisTech, France)

This paper investigates a distributed function computation setting where the un-derlying network is a rooted directed tree and where the root wants to compute afunction of the sources of information available at the nodes of the network. Themain result provides the rate region for an arbitrary function under the assumptionthat the sources satisfy a general criterion. This criterion is satisfied, in particular,when the sources are independent.

10:10

Two-Partition-Symmetrical Entropy Function RegionsQi Chen (The Chinese University of Hong Kong, Hong Kong); RaymondW. Yeung (The Chinese University of Hong Kong, Hong Kong)

Consider the entropy function region for discrete random variables Xi, i ∈ N andpartition N into N1 and N2 with 0 ≤ |N1| ≤ |N2|. An entropy function his called (N1,N2)-symmetrical if for all A,B ⊂ N , h(A) = h(B) whenever|A ∩ Ni| = |B ∩ Ni|, i = 1, 2. We prove that for |N1| = 0 or 1, the closure ofthe (N1,N2)-symmetrical entropy function region is completely characterized byShannon-type information inequalities.

10:30

Bounding the Entropic Region via Information GeometryYunshu Liu (Drexel University, USA); John M. Walsh (Drexel University,USA)

This paper suggests that information geometry may form a natural framework todeal with the unknown part of the boundary of entropic region. An application ofinformation geometry shows that distributions associated with Shannon facets canbe associated, in the right coordinates, with affine collections of distributions. Thisobservation allows an information geometric reinterpretation of the Shannon-typeinequalities as arising from a Pythagorean style relationship. The set of distribu-tions which violate Ingleton’s inequality, and hence are linked with the part of theentropic region which is yet undetermined, is shown also to have a surprising affineinformation geometric structure in a special case involving four random variablesand a certain support. These facts provide strong evidence for the link betweeninformation geometry and characterizing the boundary of the entropic region.

10:50

Characterization of the Smooth Renyi Entropy Using Ma-jorizationHiroki Koga (University of Tsukuba, Japan)

This paper unveils a new connection between majorization theory and the smoothRenyi entropy of order α. We completely characterize the subprobability distribu-

40


tion that attains the infimum included in the definition of the smooth Renyi entropyHεα(p) of order α by using the notions of majorization and the Schur convex-

ity/concavity, where p denotes a probability distribution on a discrete alphabet andε ∈ [0, 1) is an arbitrarily given constant. We can apply the obtained result to char-acterization of asymptotic behavior of 1

nHεα(p) as n → ∞ for general sources

satisfying the strong converse property.

Friday, 09:50–11:10 Room 2 – Sala de Grados

FrB2: Simulation and Approximation

Chair: Paul Cuff (Princeton University, USA)

09:50

When is it possible to simulate a DMC channel from another?Farzin Haddadpour (Sharif University of Technology, Iran); MohammadHossein Yassaee (Sharif University of Technology, Iran); MohammadReza Aref (Sharif University of Tech., Iran); Amin Aminzadeh Gohari(Sharif University of Technology, Iran)

In this paper, we study the problem of simulating a DMC channel from anotherDMC channel. We assume that the input to the channel we are simulating is i.i.d.and that the transmitter and receivers are provided with common randomness atlimited rates. We prove bounds for simulating point-to-point, MAC and broadcastchannels. As a special case, we recover the achievability part of the result of Cufffor point-to-point channel simulation via a noiseless link and shared randomness.

10:10

Joint Channel Intrinsic Randomness and Channel Resolvabil-ityAlexandre J. Pierrot (Georgia Institute of Technology, USA); Matthieu R.Bloch (Georgia Institute of Technology, France)

This paper investigates the separation of channel intrinsic randomness and channelresolvability. We derive joint exponents, which are compared to the tandem expo-nents obtained with a separate approach. We prove at once, in a simple manner,achievability results for channel intrinsic randomness, random number generation,and channel resolvability. We also provide converse results in different specialsettings.

10:30

Fixed-to-Variable Length Resolution Coding for Target Distri-butionsGeorg Bocherer (Technische Universitat Munchen, Germany); Rana AliAmjad (Technische Universitat Munchen, Germany)

The number of random bits required to approximate a target distribution in termsof un-normalized informational divergence is considered. It is shown that for avariable-to-variable length encoder, this number is lower bounded by the entropyof the target distribution. A fixed-to-variable length encoder is constructed usingM -type quantization and Tunstall coding. It is shown that the encoder achievesin the limit an un-normalized informational divergence of zero with the number ofrandom bits per generated symbol equal to the entropy of the target distribution.Numerical results show that the proposed encoder significantly outperforms theoptimal block-to-block encoder in the finite length regime.

10:50

A New Unified Method for Intrinsic Randomness Problems ofGeneral SourcesTomohiko Uyematsu (Tokyo Institute of Technology, Japan); Shohei Kuni-matsu (Tokyo Institute of Technology, Japan)

The purpose of this paper is to establish a new unified method for random numbergeneration from general sources. Specifically, we introduce an alternative defini-tion of the smooth Renyi entropy of order infinity, and show a unified approach torepresent the intrinsic randomness in terms of this information quantity. Our defi-nition of the smooth Renyi entropy is easy to calculate for finite block lengths. Wealso represent δ-intrinsic randomness and the strong converse property in terms ofthe smooth Renyi entropy.

Friday, 09:50–11:10 Room 3 – Aula 304

FrB3: Relay Networks

Chair: Michele A. Wigger (Telecom ParisTech, France)

09:50

On the Stability Region of a Relay-Assisted Multiple AccessSchemeNikolaos Pappas (Supelec, France); Marios Kountouris (Supelec,France); Anthony Ephremides (University of Maryland at College Park,USA); Apostolos Traganitis (University of Crete, Greece)

In this paper we study the impact of a relay node in a two-user network. We assumea random access collision channel model with erasures. In particular we obtain aninner and an outer bound for the stability region.

10:10

On information flow and feedback in relay networksBobbie Chern (Stanford University, USA); Ayfer Ozgur (Stanford Univer-sity, USA)

We consider wireless relay networks where a source node communicates to a des-tination node with the help of multiple intermediate relay nodes. In wireless, if anode can send information to another node, typically it can also receive informa-tion from that node. Therefore, inherently there are many possibilities for feedingback information in wireless networks. However, transmissions are not isolated butusually subject to broadcast and interference.In this paper, we ask the following question: Can the information transfer in bothdirections of a link be critical to maximizing the end-to-end communication ratein such networks? Equivalently, could one of the directions in each bidirected link(and more generally at least one of the links forming a cycle) be shut down and thecapacity of the network still be approximately maintained? Our main result is toshow that in any arbitrary Gaussian relay network with bidirected edges and cycles,we can always identify a directed acyclic subnetwork that approximately maintainsthe capacity of the original network. The edges of this subnetwork can be identifiedas the information carrying links, and the remaining links as feedback, which canonly provide limited contribution to capacity.

10:30

Two-Unicast Two-Hop Interference Network: Finite-FieldModelSong-Nam Hong (University of Southern California, USA); GiuseppeCaire (University of Southern California, USA)

In this paper we present a novel framework to convert the K-user multiple accesschannel (MAC) over Fpm into theK-user MAC over ground field Fp withmmul-tiple inputs/outputs (MIMO). This framework makes it possible to develop codingschemes for MIMO channel as done in symbol extension for time-varying chan-nels. Using aligned network diagonalization based on this framework, we showthat the sum-rate of (2m−1) log p is achievable for a 2×2×2 interference chan-nel over Fpm . We also provide some relation between field extension and symbolextension.

10:50

Cyclic Interference Neutralization on the 2×2×2 Full-DuplexTwo-Way Relay-Interference ChannelHenning Maier (RWTH Aachen University, Germany); Rudolf Mathar(RWTH Aachen University, Germany)

A two-way relay-interference channel describes a system of four communicatingtransceivers with two interjacent parallel relays arranged in a bidirectional 2× 2×2 relay-interference network. Two pairs of transceivers are each communicatingbidirectionally with the aid of both relays. All transceivers and relays are assumedto operate in full-duplex mode.Since Interference Neutralization is known as a promising method to achieve thecut-set upper bounds on the data rates of the unidirectional relay-interference chan-nel, we investigate a Cyclic Interference Neutralization scheme on the correspond-ing bidirectional relay-interference channel w.r.t. a conceptual channel model basedon a polynomial ring. We show that, if the channel matrix satisfies a certain set of

41


symmetry conditions, a total number of 4 degrees of freedom is asymptoticallyachievable.

Friday, 11:30–12:50

Friday, 11:30–12:50 Room 1 – Sala Juan Larraneta

FrC1: Error Exponent and Capacity

Chair: Tobias Koch (Universidad Carlos III de Madrid, Spain)

11:30

On the Average-Listsize Capacity and the Cutoff Rate of Dis-crete Memoryless Channels with FeedbackChristoph Bunte (ETH Zurich, Switzerland); Amos Lapidoth (ETHZ,Switzerland)

We study the cutoff rate and the average-listsize capacity of discrete memorylesschannels (DMCs) with feedback. We show that feedback can increase the average-listsize capacity but not the cutoff rate. For DMCs with positive zero-error capacity,we show that the average-listsize capacity with feedback is equal to the cutoff rate.For all other DMCs, we derive a lower bound on the average-listsize capacity withfeedback. The bound is asymptotically tight for low-noise channels. We also showthat a multi-letter version of Forney’s lower bound on the average-listsize capacityof DMCs without feedback is asymptotically tight.

11:50

Improved Capacity Lower Bounds for Channels with Dele-tions and InsertionsRamji Venkataramanan (University of Cambridge, United Kingdom);Sekhar Tatikonda (Yale University, USA)

New lower bounds are obtained for the capacity of a binary channel with deletionsand insertions. Each input bit to the channel is deleted with probability d, or anextra bit is inserted after it with probability i, or it is transmitted unmodified withprobability 1 − d − i. This paper builds on the idea introduced in [1] of using asub-optimal decoder that decodes the positions of the deleted and inserted runs, inaddition to the transmitted codeword. The mutual information between the channelinput and output sequences is expressed as the sum of the rate achieved by thisdecoder and the rate loss due to its suboptimality. The main contribution is ananalytical lower bound for the rate loss term which leads to an improvement in thecapacity lower bound of [1]. For the special case of the deletion channel, the newbound is larger than the previous best lower bound for deletion probabilities up to0.3.

12:10

ε-Capacity and Strong Converse for Channels with GeneralStateMarco Tomamichel (National University of Singapore, Singapore); Vin-cent Y. F. Tan (Institute for Infocomm Research, Singapore)

We consider state-dependent memoryless channels with general state availableat both encoder and decoder. We establish the ε-capacity and the optimistic ε-capacity. This allows us to prove a necessary and sufficient condition for the strongconverse to hold. We also provide a simpler sufficient condition on the first- andsecond-order statistics of the state process that ensures that the strong converseholds.

Friday, 11:30–12:50 Room 2 – Sala de Grados

FrC2: Topics in Compression

Chair: Claudio Weidmann (CNRS / ENSEA / University Cergy-Pontoise,France)

11:30

Compression of Noisy Signals with Information BottlenecksAmin Emad (University of Illinois at Urbana-Champaign, USA); Olgica

Milenkovic (University of Illinois, USA)

We consider a novel approach to the information bottleneck problem where the goalis to perform compression of a noisy signal, while retaining a significant amountof information about a correlated auxiliary signal. To facilitate analysis, we castcompression with side information as an optimization problem involving an infor-mation measure, which for jointly Gaussian random variables equals the classicalmutual information. We provide closed form expressions for locally optimal lin-ear compression schemes; in particular, we show that the optimal solutions are ofthe form of the product of an arbitrary full-rank matrix and the left eigenvectorscorresponding to smallest eigenvalues of a matrix related to the signals’ covariancematrices. In addition, we study the influence of the sparsity level of the Bernoulli-Gaussian noise on the compression rate. We also highlight the similarities anddifferences between the noisy bottleneck problem and canonical correlation analy-sis (CCA), as well as the Gaussian information bottleneck problem.

11:50

Distortion Minimization in Layered Broadcast Transmissionof a Gaussian Source Over Rayleigh ChannelsWessam Mesbah (King Fahd University of Petroleum and Minerals, SaudiArabia); Mohammad Shaqfeh (Texas A&M University at Qatar, Qatar);Hussein Alnuweiri (Texas A&M University, Qatar)

We consider the problem of minimizing the expected distortion in the multilayertransmission of a Gaussian source using the broadcast approach with successiveinformation enhancement. This minimization is contingent on the jointly optimalchoice of the rates and power ratios of the different layers. This problem was tack-led in the literature with the assumption that the fading channel has a finite numberof states and the number of source layers matches the number of channel states. Inthis paper, we provide a more generic solution for a continuous Rayleigh fadingchannel, and for any predetermined number of layers. We prove that the primaloptimization problem has a strong duality with the Lagrangian dual problem. Con-sequently, we propose a two-dimensional bisection search algorithm that can, forany number of layers, find the optimal solution of the dual problem which will bethe same as the optimal solution of the primal problem. The complexity of thesearch algorithm has linear order with respect to the number of layers. We providenumerical results for the optimal rate and power allocation. Moreover, we showthat with a small number of layers, we can approach the distortion lower boundthat is achieved by transmitting an infinite number of layers.

12:10

Sparse Signal Processing with Linear and Non-Linear Obser-vations: A Unified Shannon Theoretic ApproachCem Aksoylar (Boston University, USA); George Atia (University of Cen-tral Florida, USA); Venkatesh Saligrama (Boston University, USA)

In this work we derive fundamental limits for many linear and non-linear sparsesignal processing models including group testing, quantized compressive sensing,multivariate regression and observations with missing features. In general sparsesignal processing problems can be characterized in terms of the following Marko-vian property. We are given a set of N variables X1, X2, . . . , XN , and thereis an unknown subset S ⊂ 1, 2, . . . , N that are relevant for predicting out-comes/outputs Y . In other words, when Y is conditioned on Xnn∈S it is con-ditionally independent of the other variables, Xnn 6∈S . Our goal is to identifythe set S from samples of the variables X and the associated outcomes Y . Wecharacterize this problem as a version of the noisy channel coding theorem. Usingasymptotic information theoretic analyses, we describe mutual information for-mulas that provide sufficient and necessary conditions on the number of samplesrequired to successfully recover the salient variables. This mutual information ex-pression unifies conditions for both linear and non-linear observations. We thencompute sample complexity bounds for the aforementioned models, based on themutual information expressions.

12:30

Rate-Distortion Bounds for an Epsilon-Insensitive DistortionMeasureKazuho Watanabe (Nara Institute of Science and Technology, Japan)

Direct evaluation of the rate-distortion function has rarely been achieved when itis strictly greater than its Shannon lower bound. In this paper, we consider therate- distortion function for the distortion measure defined by an ε-insensitive lossfunction. We first present the Shannon lower bound applicable to any source distri-

42


bution with finite differential entropy. Then, focusing on the Laplacian and Gaus-sian sources, we prove that the rate-distortion functions of these sources are strictlygreater than their Shannon lower bounds and obtain analytic upper bounds for therate-distortion functions. Small distortion limit and numerical evaluation of thebounds suggest that the Shannon lower bound provides a good approximation tothe rate-distortion function for the ε-insensitive distortion measure.

Friday, 11:30–12:50 Room 3 – Aula 304

FrC3: Capacity of Relay Networks

Chair: Nikolaos Pappas (Supelec, France)

11:30

Improved Capacity Approximations for Gaussian Relay Net-worksRitesh Kolte (Stanford University, USA); Ayfer Ozgur (Stanford Univer-sity, USA)

Consider a Gaussian relay network where a number of sources communicate toa destination with the help of several layers of relays. Recent work has shownthat a compress-and-forward based strategy at the relays can achieve the capacityof this network within an additive gap. In this strategy, the relays quantize theirobservations at the noise level and map it to a random Gaussian codebook. Theresultant gap is independent of the SNR’s of the channels in the network but linearin the total number of nodes.In this paper, we show that if the relays quantize their signals at a resolution de-creasing with the number of nodes in the network, the additive gap to capacity canbe made logarithmic in the number of nodes for a class of layered, time-varyingwireless relay networks. This suggests that the rule-of-thumb to quantize the re-ceived signals at the noise level used for compress-and- forward in the currentliterature can be highly suboptimal.

11:50

Gaussian Half-Duplex Relay Networks: Improved Gap and aConnection with the Assignment ProblemMartina Cardone (Eurecom, France); Daniela Tuninetti (University ofIllinois at Chicago, USA); Raymond Knopp (Institut Eurecom, France);Umer Salim (Intel Mobile Communications, France)

This paper studies a Gaussian relay network, where the relays can either transmitor receive at any given time, but not both. Known upper (cut-set) and lower (noisynetwork coding) bounds on the capacity of a memoryless full-duplex relay net-work are specialized to the half-duplex case and shown to be to within a constantgap of one another. For fairly broad range of relay network sizes, the derived gapis smaller than what is known in the literature, and it can be further reduced formore structured networks such as diamond networks. It is shown that the asymp-totically optimal duration of the listen and transmit phases for the relays can beobtained by solving a linear program; the coefficients of the linear constraints ofthis linear program are the solution of certain ‘assignment problems’ for which ef-ficient numerical routines are available; this gives a general interesting connectionbetween the high SNR approximation of the capacity of a MIMO channel and the‘assignment problem’ in graph theory. Finally, some results available for diamondnetworks are extended to general networks. For a general relay network with 2relays, it is proved that, out of the 4 possible listen/transmit states, at most 3 havea strictly positive probability. Numerical results for a network with K − 2 < 9relays show that at most K − 1 states have a strictly positive probability, which isconjectured to be true for any number of relays.

12:10

Capacity Region of a Class of Interfering Relay ChannelsHieu T. Do (Royal Institute of Technology (KTH), Sweden); TobiasJ. Oechtering (KTH Royal Institute of Technology, Sweden); MikaelSkoglund (KTH Royal Institute of Technology, Sweden); Mai Vu (TuftsUniversity, USA)

This paper studies a new model for cooperative communication, the interfering re-lay channels. We show that the hash-forward scheme introduced by Kim for theprimitive relay channel is capacity achieving for a class of semideterministic inter-fering relay channels. The obtained capacity result generalizes and unifies earlier

capacity results for a class of primitive relay channels and a class of deterministicinterference channels.

12:30

The Deterministic Capacity of Relay Networks with Relay Pri-vate MessagesAhmed A. Zewail (Nile University, Egypt); Yahya Mohasseb (Nile Uni-versity, Egypt); Mohammed Nafie (Nile University, Egypt); Hesham ElGamal (Ohio State University, USA)

We study the capacity region of a deterministic 4-node network, where 3 nodes canonly communicate via the fourth one. However, the fourth node is not merely arelay since it can exchange private messages with all other nodes. This situationresembles the case where a base station relays messages between users and deliversmessages between the backbone system and the users. We assume an asymmetricscenario where the channel between any two nodes is not reciprocal. First, an upperbound on the capacity region is obtained based on the notion of single sided genie.Subsequently, we construct an achievable scheme that achieves this upper boundusing a superposition of broadcasting node 4 messages and an achievable “detour”scheme for a reduced 3-user relay network.

43


Author IndexAAbbe, Emmanuel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Afser, Huseyin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Agustin, Adrian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Ahmadian, Mahmoud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Aihara, Kazuyuki . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Akhbari, Bahareh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Aksoylar, Cem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42Akyol, Emrah . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Alkhouli, Tamer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Alnuweiri, Hussein . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42Alon, Noga . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Alsan, Mine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Amjad, Rana Ali . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .41Anantharam, Venkat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32Aref, Mohammad Reza . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34, 41Asadi, Meysam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37Asnani, Himanshu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Atia, George . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42Avestimehr, Salman . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Awan, Zohaib Hassan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Ayday, Erman . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

BBadr, Arash Shahbaz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Bai, Bao-Ming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27, 33Bakshi, Mayank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Banerjee, Siddhartha . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Barbier, Jean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Barron, Andrew R. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Bazarsky, Alex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Beck, Eugen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Behroozi, Hamid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36Belfiore, Jean-Claude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Benammar, Meryem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Bernal, Jose Joaquın . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Bhattacharya, Chiranjib . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Bitouze, Nicolas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Matthieu R., Matthieu R. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Bloch, Matthieu R. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Boche, Holger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Bocherer, Georg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Bossert, Martin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Brito, Cesar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Bueno-Carreno, Diana . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Bunte, Christoph. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .42Burnashev, Marat V. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

CCai, Sheng . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Caire, Giuseppe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26, 41Calderbank, Robert . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

Campello, Antonio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Cardone, Martina . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Cesa-Bianchi, Nicolo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Chakraborty, Supriyo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Chan, Chun Lam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Chan, Chung . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Chan, Terence H. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Chen, Jun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Chen, Qi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Chen, Xu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Chern, Bobbie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Chou, Remi A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Cohen, Asaf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Costa, Sueli I. R. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29Costello, Jr., Daniel J. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25Crespo, Pedro M. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Cuff, Paul . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24, 35, 39Czap, Laszlo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

DDas, Abhik Kumar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Das, Niladri . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Dean, Thomas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Delic, Hakan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Devroye, Natasha . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32Diggavi, Suhas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24, 32Dimakis, Alexandros . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Do, Hieu T. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Dolecek, Lara . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23, 37Duan, Ruchen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36Dubhashi, Devdatt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Duchi, John . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

EEffros, Michelle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30, 31, 38Einolghozati, Arash . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25El Gamal, Hesham . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43El-Khatib, Rafah. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25Emad, Amin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37, 42Ephremides, Anthony . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34, 41Erez, Uri . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Ernvall, Toni . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

FFarnoud (Hassanzadeh), Farzad . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25Feder, Meir . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Fekri, Faramarz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25Fernandes, Winston . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Ferreira, Hendrik C. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Fitzek, Frank H.P. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .24Fraenkel, Ernest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Fragouli, Christina . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

44


GGan, Lu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Garivier, Aurelien. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37Geiger, Bernhard C. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Geng, Chunhua . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Gentile, Claudio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Ghasemi-Goojani, Shahab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36Girnyk, Maksym A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Gohari, Amin Aminzadeh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Goldsmith, Andrea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22, 26, 34Grant, Alex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Gripon, Vincent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37Guan, Yong Liang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Guo, Dongning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Gurewitz, Omer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Gutierrez-Gutierrez, Jesus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Gyorfi, Laszlo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

HHaddadpour, Farzin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Hassani, S. Hamed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Hassani, S. Hamed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Heymann, Reolyn. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .33Ho, Tracey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Hong, Song-Nam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Høst-Madsen, Anders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Huang, Sheng . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Huang, Yu-Chih . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

IIshwar, Prakash . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

JJafar, Syed Ali . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Jaggi, Sidharth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Jayram, T. S. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Jee Lee, Eun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Jethava, Vinay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Ji, Mingyue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Jitsumatsu, Yutaka . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Jordan, Michael . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

KKampeas, Joseph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Karagoz, Batuhan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Karbasi, Amin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Karl, William. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .27Kashyap, Navin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31, 38Kearns, Michael . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Khisti, Ashish . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23, 36Khoshnevisan, Mostafa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Kim, Minji . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37Kliewer, Joerg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Klotz, Johannes Georg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Knopp, Raymond . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Kobayashi, Mari . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Koga, Hiroki . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Kohda, Tohru . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .35Kolte, Ritesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Kontoyiannis, Ioannis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Kostina, Victoria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Kountouris, Marios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34, 41Koyluoglu, Onur Ozan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Krzakala, Florent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Kumar Rai, Brijesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Kunimatsu, Shohei . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

LLabeau, Fabrice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Laneman, J. Nicholas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Langberg, Michael . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30, 31Lapidoth, Amos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32, 42Le Treust, Mael . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .22Lelarge, Marc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22, 29, 39Levi, Ilia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Li, Minghua . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Li, Yao . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Li, Ying . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Liang, Yingbin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36Ligo, Jonathan. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37Ling, Cong . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Litsyn, Simon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Liu, Jia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Liu, Wei . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Liu, Yang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Liu, Yunshu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Lopez-Martinez, Francisco Javier . . . . . . . . . . . . . . . . . . . . . . . . . . 26Loyka, Sergey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Lucani, Daniel E. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Lugosi, Gabor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36Luzzi, Laura . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

MMa, Xiao . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27, 33Maamari, Diana . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32Macris, Nicolas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25Madiman, Mokshay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Maier, Henning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Manickam, Shivkumar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Mansour, Yishay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Martos-Naya , Eduardo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Mashiach, Adam. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .38Massoulie, Laurent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Mathar, Rudolf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Mehmetoglu, Mustafa S. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Merhav, Neri . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Mesbah, Wessam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42Milenkovic, Olgica . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25, 37, 42

45


Mili, Lamine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Miloslavskaya, Vera . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .21Mitchell, David G. M. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25Mohasseb, Yahya . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Molisch, Andreas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Mondelli, Marco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Muramatsu, Jun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

NNafie, Mohammed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Narayanan, Krishna . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30, 31Nasiraee, Mohammad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Ney, Hermann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Nokleby, Matthew . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Nussbaum-Thom, Markus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

OOechtering, Tobias J. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Oggier, Frederique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Olmos, Pablo M.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25Ordentlich, Or . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Orten, Birant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Østergaard, Jan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Ozgur, Ayfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41, 43

PPalguna, Deepan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Pappas, Nikolaos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34, 41Parada, Patricio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Paris, Jose Francisco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Pedersen, Morten V. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .24Pei ying, Zhang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Pfister, Henry D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Piantanida, Pablo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26, 40Pierrot, Alexandre J. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Pitaval, Renaud-Alexandre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .33Pollak, Ilya . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .35Prabhakaran, Vinod M. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Presman, Noam. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .27Proutiere, Alexandre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

RRabbat, Michael . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37Raginsky, Maxim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Rakhlin, Alexander . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Rasmussen, Lars K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Rini, Stefano . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Rodrigues, Miguel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Rohban, Mohammad Hossein . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Rondao Alface, Alface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Rose, Kenneth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

SSørensen, Chres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

Sabeti, Elyas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Sabharwal, Ashutosh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32Sahai, Achaleshwar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32Sala, Frederic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37Salavati, Amir Hesam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Saligrama, Venkatesh. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .26, 27, 42Salim, Umer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Santhanam, Narayana Prasad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37Savard, Anne . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Schaefer, Rafael F. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Schieler, Curt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .39Schlegel, Christian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25Schluter, Ralf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .35Schober, Steffen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Schouhamer Immink, Kees A.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37Sefidgaran, Milad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Shamai, Shlomo (Shitz) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24, 36Shao, Shuo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Shaqfeh, Mohammad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42Shimonovich, Jonathan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36Shokrollahi, Amin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29, 36Shomorony, Ilan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Shum, Kenneth W. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Si, Hongbo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Silva, Jorge F. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Simon, Juan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Skoglund, Mikael . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38, 43Soljanin, Emina . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Somekh-Baruch, Anelia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36Song, Eva Chen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Song, Lin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Sridharan, Karthik . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Srivastava, Mani B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Sun, Hua . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Sun, Qifu T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Swart, Theo G. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33, 37Szczecinski, Leszek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Sznajdman, Jacob . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

TTakeuchi, Junichi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Talebi, M. Sadegh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Tan, Louis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Tan, Vincent Y. F. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42Tang, Siyu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23Tatikonda, Sekhar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37, 42Taylor, Jr., Robert M. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29Tchamkerten, Aslan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Tehrani, Arash Saber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26Temmel, Christoph. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29Thakor, Satyajit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22, 38Timo, Roy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Tirkkonen, Olav . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Tishby, Naftali . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .32

46


Tomamichel, Marco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42Torghabeh, Ramezan Paravi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37Toumpakaris, Dimitris . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .39Traganitis, Apostolos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .41Trifonov, Peter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Truhachev, Dmitri . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25Tunali, Nihat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Tuninetti, Daniela . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32, 43Tutuncuoglu, Kaya . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

UUrbanke, Rudiger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .21, 24, 25, 38Uyematsu, Tomohiko . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

VVaishampayan, Vinay A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Vandendorpe, Luc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Veeravalli, Venugopal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37Vehkalahti, Roope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Vehkapera, Mikko . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Vellambi, Badri N. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Venkataramanan, Ramji . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37, 42Verdu, Sergio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34Vidal, Josep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Vilenchik, Danny . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Vishwanath, Sriram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21, 30Vu, Mai . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

WWainwright, Martin J. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21, 39Walsh, John Maclaren . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Walton, Chad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Warsi, Naqueeb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35Watanabe, Kazuho . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42Weber, Jos H. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Weidmann, Claudio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Weissman, Tsachy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22Wiese, Moritz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Wigger, Michele . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Wu, Youlong . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

XXu, Jiaming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Xu, Xiaoli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

YYassaee, Mohammad Hossein . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Yavuz, Semih . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Yener, Aylin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31Yeung, Raymond W. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Yu, Wei . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

ZZaghloul, Amir . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Zaidi, Abdellatif . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Zamir, Ram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Zdeborova, Lenka . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Zewail, Ahmed A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43Zhang, Pan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Zhou, Yuhan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Zhuang, Qiutao . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33Zou, Jun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

47

NOTES

48

NOTES

49

NOTES

50

NOTES

51

NOTES

52

2013 IEEE Information Theory Workshopitw2013.tsc.uc3m.es/sites/itw2013.tsc.uc3m.es/files/... ·...

Documents

Transcript of 2013 IEEE Information Theory Workshopitw2013.tsc.uc3m.es/sites/itw2013.tsc.uc3m.es/files/... ·...