Networked CPS: Some Fundamental Challenges 

John S. BarasInstitute for Systems Research

Department of Electrical and Computer EngineeringFischell Department of BioengineeringDepartment of Mechanical Engineering

Applied Mathematics, Statistics and Scientific Computation ProgramUniversity of Maryland College Park

Panel on “Networking challenges for CPS”2014 INFOCOM

Toronto, Canada, May 1, 2014

Wireless and Networked Embedded Systems: Ubiquitous Presence

2

A Network Immersed World: Swarms and the Cloud

Courtesy: J. Rabaey3

Networked CPS: Wireless Sensor Networks  Everywhere

4

Networked CPS: Smart Grids

Courtesy: Rockwell 5

Networked CPS: FAA NextGen

6

At work: Two ASIMOs working together in coordination to deliver refreshments

Credit: Honda

• Component-based Architectures• Communication vs Performance

Tradeoffs• Distributed asynchronous• Fundamental limits

Networked CPS: Autonomous Swarms

7

A Network Immersed World

• A complex collection of sensors, controllers, computing nodes, and actuators that work together to improve our daily lives– From very small: Ubiquitous, Pervasive, Disappearing, perceptive, Ambient

– To very large: Always Connectable, Reliable, Scalable, Adaptive, Flexible

• Emerging Service Models– Building energy management– Automotive safety and control– Management of metropolitan traffic flows– Distributed health monitoring– Smart Grid

8

• Multiple Interacting Graphs – Nodes: agents, individuals, groups, 

organizations– Directed graphs– Links: ties, relationships– Weights on links : value (strength, 

significance) of tie– Weights on nodes : importance of node (agent)

• Value directed graphs with weighted nodes

• Real‐life problems: Dynamic, time varying graphs,  relations, weights, policies

Information network

Communication network

Sijw : S

ii w

: Sjj w

Iklw: I

kk w : Ill w

Cmnw: C

mm w : Cnn w

Networked System architecture & operation

Networked CPS Architecture: Multiple Interacting Multigraphs

9

Networks: The Fundamental Trade‐off

• The nodes gain from collaborating• But collaboration has costs (e.g. communications)• Trade‐off: gain from collaboration vs cost of  

collaborationVector metrics involved typicallyConstrained Coalitional Games

10

Example 1: Network Formation  ‐‐ Effects on Topology Example 2: Collaborative robotics, communications Example 3: Web‐based social networks and services

● ● ●

Example 4: Groups of cancer tumor or virus cells

11

Simple Lattice C(n,k) Small world: Slight variation adding 

Efficient Communication Graphs:Small World Graphs

nk

Adding a small portion of well‐chosen links → significant increase in convergence rate

Efficient Communication Graphs: Expander Graphs

• Fast synchronization of a network of oscillators • Network where any node is “nearby” any other • Fast ‘diffusion’ of information in a network• Fast convergence of consensus  • Decide connectivity with smallest memory • Random walks converge rapidly• Easy to construct, even in a distributed way (ZigZag graph product)

• Graph G,  Cheeger constant h(G)– All partitions of G to S and Sc , 

h(G)=min (#edges connecting S and Sc )  / (#nodes in smallest of S and Sc )

• (k , N, e) expander : h(G) > e ; sparse but locally well connected  (1‐SLEM(G) increases as h(G)2)

12

Expander Graphs – Ramanujan Graphs

13

14

Examples of resulting topologies

Construction of Efficient Communication Graphs by Computational Optimization

15

Distributed self ‐ organization

Goal: design a scheme that gives each node a vector of compact global information

• Component Based Networking : Leads to a compositional approach to the synthesis and operation of networks. Fits very well MANET, WNAN (and WAND), beyond.

• Does away with classical layers and with classical cross‐layer

• Compositionality, and Compositional Synthesis

• Cross linked executable, formal and performance models is addressing this challenging problem directly.

Component Based Networking and Security

16

Interacting Control, Information and Communication Graphs

Component‐base Networks and Composable Security 

Executable Models

Performance Models

Formal Models

Universally ComposableSecurity of Network Protocols:• Network with many agents running

autonomously. • Agents execute in mostly asynchronous

manner, concurrenty several protocols many times. Protocols may or may have not been jointly designed, may or not be all secure or secure to same degree.

Key question addressed : • Under what conditions can the

composition of these protocols be provably secure?

• Investigate time and resource requirements for achieving this

Studying compositionality is necessary!

Compositional Security is critical for all CPS!

17

18

Trust and Collaborative Control/Operation

Two linked dynamics • Trust / Reputation propagation      

and collaborative control evolution

• Beyond linear algebra and weights, semirings of constraints, constraint programming, soft constraints semirings, policies, agents

• Learning on graphs and network dynamic games: behavior, adversaries• Adversarial models, attacks, constrained shortest paths, …

• Integrating network utility maximization (NUM) with constraint based reasoning and coalitional games

Interacting Control, Information and Communication Graphs

Biological Network Types

19

Examples of biological networks: [A] Yeast transcription factor‐binding network; [B] Yeast protein‐protein interaction network; [C] Yeast phosphorylation network ; [D] E. Coli metabolic network ; [E] Yeast genetic network ;  Nodes colored according to their YPD cellular roles [Zhu et al, 2007]

How Biology Does IT?

20

Modularity vs Performance

• Optimize only on performance – poor adaptivity• Add cost of communications – improved adaptivity• Communication motifs• Evolvable modularity for some networked CPS??

21

Neural Network Evolution:from programmed structure to function 

feedback on structure

22

Social Networks as Networked CPS

• We are much more “social” than ever before– Online social networks (SNS) permeate our lives– Such new Life style gives birth to new markets

• Monetize the value of social network– Advertising - major source of income for SNS– Joining fee, donation etc.– …

• Need to know the common featuresof social networks

23

Challenges in Social Networks

24

Can we integrate? Context‐based distribution Include user and product similarity Combine with user‐user similarity Exploit both user preferences and network structure  Maximize relevance and potential profit  Ensure message delivery to all interested nodes Increase recommendation accuracy and diversity

Can hyperbolic embedding help? 

Is it real?

Key Idea: Virtual Geometry

• Of the network graph • Of an auxiliary space underlying the graph

25

Possible Underlying Hyperbolic Geometry?

26

27

Navigation Efficiency and Robustness

• Percentage of successful greedy paths 99.99%

• Percentage of shortest greedy paths 100%

• Percentage of successful greedy paths after removal of x% of links or nodes– x =10% 99%– x =30% 95%

Taxonomy for Large‐Scale Networks based on Curvature

Courtesy: Iraj Saniee, Bell Labs 28

1987 – Gromov: d ‐ hyperbolicity

Thank you!

baras@umd.edu301‐405‐6606

http://www.isr.umd.edu/~baras

Questions?