Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark...
-
Upload
trinhthuan -
Category
Documents
-
view
218 -
download
1
Transcript of Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark...
![Page 1: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/1.jpg)
Transfer Learning with Applications
Sinno Jialin Pan 1, Qiang Yang2,3 and Wei Fan3
1 Institute for Infocomm Research, Singapore 2 Hong Kong University of Science and Technology
3 Huawei Noah's Ark Research Lab, Hong Kong
![Page 2: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/2.jpg)
Outline
Part I: An overview of transfer learning – (Sinno J. Pan) Part II: Transfer learning applications (Prof. Qiang Yang)
Part III: Advanced research topics: heterogeneous transfer
learning (Wei Fan)
2
![Page 3: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/3.jpg)
Transfer Learning Overview
Sinno Jialin Pan (Ph.D.) Lab Head, Text Analytics,
Data Analytics Department, Institute for Infocomm Research (I2R), Singapore
![Page 4: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/4.jpg)
Transfer of Learning A psychological point of view
• The study of dependency of human conduct, learning or performance on prior experience. – [Thorndike and Woodworth, 1901] explored how individuals
would transfer in one context to another context that share similar characteristics.
C++ Java Maths/Physics Computer Science/Economics
2
![Page 5: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/5.jpg)
Transfer Learning In the machine learning community
• The ability of a system to recognize and apply knowledge and skills learned in previous domains/tasks to novel tasks/domains, which share some commonality.
• Given a target domain/task, how to identify the commonality between the domain/task and previous domains/tasks, and transfer knowledge from the previous domains/tasks to the target one?
3
![Page 6: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/6.jpg)
Transfer Learning
4
Traditional Machine Learning Transfer Learning
train
ing
dom
ains
test
dom
ains
train
ing
item
s
test
item
s
domain A domain B domain C
![Page 7: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/7.jpg)
Transfer Learning Different fields
• Transfer learning for reinforcement learning.
[Taylor and Stone, Transfer Learning for Reinforcement Learning Domains: A Survey, JMLR 2009]
• Transfer learning for classification, and regression problems.
[Pan and Yang, A Survey on
Transfer Learning, IEEE TKDE 2010]
5
Focus!
![Page 8: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/8.jpg)
Motivating Example I: Indoor WiFi localization
6
-30dBm -70dBm -40dBm
![Page 9: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/9.jpg)
Indoor WiFi Localization (cont.)
7
Training
Training Test
Device A
Test
Device B
~ 1.5 meters
~10 meters
Device A
Device A
S=(-37dbm, .., -77dbm), L=(1, 3) S=(-41dbm, .., -83dbm), L=(1, 4) … S=(-49dbm, .., -34dbm), L=(9, 10) S=(-61dbm, .., -28dbm), L=(15,22)
S=(-37dbm, .., -77dbm) S=(-41dbm, .., -83dbm) … S=(-49dbm, .., -34dbm) S=(-61dbm, .., -28dbm)
S=(-37dbm, .., -77dbm) S=(-41dbm, .., -83dbm) … S=(-49dbm, .., -34dbm) S=(-61dbm, .., -28dbm)
S=(-33dbm, .., -82dbm), L=(1, 3) … S=(-57dbm, .., -63dbm), L=(10, 23)
Localization model
Localization model
Drop!
Average Error Distance
![Page 10: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/10.jpg)
Difference between Domains
8
Time Period A Time Period B
Device B
Device A
![Page 11: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/11.jpg)
Motivating Example II: Sentiment classification
9
![Page 12: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/12.jpg)
Sentiment Classification (cont.)
10
Training
Training Test
Electronics
Test
~ 84.6%
~72.65%
Sentiment Classifier
Sentiment Classifier
Drop! Electronics
Classification Accuracy
Electronics DVD
![Page 13: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/13.jpg)
Difference between Domains
11
Electronics Video Games (1) Compact; easy to operate; very good picture quality; looks sharp!
(2) A very good game! It is action packed and full of excitement. I am very much hooked on this game.
(3) I purchased this unit from Circuit City and I was very excited about the quality of the picture. It is really nice and sharp.
(4) Very realistic shooting action and good plots. We played this and were hooked.
(5) It is also quite blurry in very dark settings. I will never buy HP again.
(6) The game is so boring. I am extremely unhappy and will probably never buy UbiSoft again.
![Page 14: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/14.jpg)
A Major Assumption in Traditional Machine Learning Training and future (test) data come from
the same domain, which implies Represented in the same feature spaces.
Follow the same data distribution.
12
![Page 15: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/15.jpg)
In Real-world Applications
• Training and testing data may come from different domains, which have: Different marginal distributions, or different
feature spaces: Different predictive distributions, or different
label spaces:
13
![Page 16: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/16.jpg)
How to Build Systems on Each Domain of Interest
Build every system from scratch? Time consuming and expensive!
Reuse common knowledge extracted from
existing systems? More practical!
14
![Page 17: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/17.jpg)
The Goal of Transfer Learning
15
Source Domain Data
Target Domain Data
Predictive Models
Labeled Training
Unlabeled data/a few labeled data for adaptation
Transfer Learning Algorithms
Target Domain Data
Testing
Electronics
Time Period A
Device A
DVD Device B Time Period B
![Page 18: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/18.jpg)
Transfer Learning Settings
16
Transfer Learning
Heterogeneous Transfer Learning
Heterogeneous
Feature Space
Homogeneous Transfer Learning
Homogeneous Unsupervised Transfer
Learning
Semi-Supervised Transfer Learning
Supervised Transfer Learning
![Page 19: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/19.jpg)
Transfer Learning Approaches
17
Instance-based Approaches
Feature-based Approaches
Parameter-based Approaches
Relational Approaches
![Page 20: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/20.jpg)
Instance-based Transfer Learning Approaches
Source and target domains have a lot of overlapping features (domains share the same/similar support)
18
General Assumption
![Page 21: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/21.jpg)
Instance-based Transfer Learning Approaches
Case I
Case II
19
Problem Setting
Assumption Assumption
Problem Setting
![Page 22: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/22.jpg)
Instance-based Approaches Case I
Given a target task,
20
![Page 23: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/23.jpg)
Instance-based Approaches Case I (cont.)
21
![Page 24: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/24.jpg)
Instance-based Approaches Case I (cont.)
Assumption:
22
![Page 25: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/25.jpg)
Instance-based Approaches Case I (cont.)
23
Correcting Sample Selection Bias / Covariate Shift [Quionero-Candela, etal, Data Shift in Machine Learning, MIT Press 2009]
![Page 26: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/26.jpg)
Assumption: sample selection bias is caused by the data generation process
Instance-based Approaches Correcting sample selection bias
• Imagine a rejection sampling process, and view the source domain as samples from the target domain
24
![Page 27: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/27.jpg)
Instance-based Approaches Correcting sample selection bias (cont.)
• The distribution of the selector variable maps the target onto the source distribution
25
Label instances from the source domain with label 1 Label instances from the target domain with label 0 Train a binary classifier
[Zadrozny, ICML-04]
![Page 28: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/28.jpg)
Instance-based Approaches Kernel mean matching (KMM)
Maximum Mean Discrepancy (MMD) [Alex Smola, Arthur Gretton and Kenji Kukumizu, ICML-08 tutorial]
26
![Page 29: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/29.jpg)
Instance-based Approaches Kernel mean matching (KMM) (cont.)
27
[Huang etal., NIPS-06]
![Page 30: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/30.jpg)
Instance-based Approaches Direct density ratio estimation
28
[Sugiyama etal., NIPS-07, Kanamori etal., JMLR-09]
KL divergence loss Least squared loss
[Sugiyama etal., NIPS-07] [Kanamori etal., JMLR-09]
![Page 31: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/31.jpg)
Instance-based Approaches Case II
• Intuition: Part of the labeled data in the source domain can be reused in the target domain after re-weighting
29
![Page 32: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/32.jpg)
Instance-based Approaches Case II (cont.)
TrAdaBoost [Dai etal ICML-07] – For each boosting iteration, Use the same strategy as AdaBoost to
update the weights of target domain data. Use a new mechanism to decrease the
weights of misclassified source domain data.
30
![Page 33: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/33.jpg)
Feature-based Transfer Learning Approaches
When source and target domains only have some overlapping features. (lots of features only have support in either the source or the target domain)
31
![Page 34: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/34.jpg)
Feature-based Transfer Learning Approaches (cont.)
How to learn ? Solution 1: Encode application-specific
knowledge to learn the transformation. Solution 2: General approaches to learning
the transformation.
32
![Page 35: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/35.jpg)
Feature-based Approaches Encode application-specific knowledge
33
Electronics Video Games (1) Compact; easy to operate; very good picture quality; looks sharp!
(2) A very good game! It is action packed and full of excitement. I am very much hooked on this game.
(3) I purchased this unit from Circuit City and I was very excited about the quality of the picture. It is really nice and sharp.
(4) Very realistic shooting action and good plots. We played this and were hooked.
(5) It is also quite blurry in very dark settings. I will never_buy HP again.
(6) The game is so boring. I am extremely unhappy and will probably never_buy UbiSoft again.
![Page 36: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/36.jpg)
Feature-based Approaches Encode application-specific knowledge (cont.)
34
compact sharp blurry hooked realistic boring 1 1 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0
( ) sgn( ), [1,1, 1,0,0,0]Ty f x w x w= = ⋅ = −
compact sharp blurry hooked realistic boring 0 0 0 1 0 0 0 0 0 1 1 0 0 0 0 0 0 1
Electronics
Video Game
Training
Prediction
![Page 37: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/37.jpg)
Feature-based Approaches Encode application-specific knowledge (cont.)
35
Electronics Video Games (1) Compact; easy to operate; very good picture quality; looks sharp!
(2) A very good game! It is action packed and full of excitement. I am very much hooked on this game.
(3) I purchased this unit from Circuit City and I was very excited about the quality of the picture. It is really nice and sharp.
(4) Very realistic shooting action and good plots. We played this and were hooked.
(5) It is also quite blurry in very dark settings. I will never_buy HP again.
(6) The game is so boring. I am extremely unhappy and will probably never_buy UbiSoft again.
![Page 38: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/38.jpg)
Feature-based Approaches Encode application-specific knowledge (cont.)
Three different types of features Source domain (Electronics) specific features, e.g., compact, sharp, blurry Target domain (Video Game) specific features, e.g., hooked, realistic, boring Domain independent features (pivot features), e.g., good, excited, nice, never_buy
36
![Page 39: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/39.jpg)
Feature-based Approaches Encode application-specific knowledge (cont.)
How to identify pivot features? Term frequency on both domains Mutual information between features and labels (source domain) Mutual information on between features and domains
How to utilize pivots to align features across domains? Structural Correspondence Learning (SCL) [Biltzer etal.
EMNLP-06] Spectral Feature Alignment (SFA) [Pan etal. WWW-10]
37
![Page 40: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/40.jpg)
Feature-based Approaches Structural Correspondence Learning (SCL)
Intuition Use pivot features to construct pseudo tasks that
related to target classification task Model correlations between pivot features and
other features using multi-task learning techniques Discover new shared features by exploiting the
feature correlations
38
![Page 41: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/41.jpg)
Structural Correspondence Learning Algorithm
Identify P pivot features Build P classifiers to predict the pivot features
from remaining features Discover shared feature subspace Compute top K eigenvectors Project original features into eigenvectors to
derive new shared features Train classifiers on the source using augmented
features (original features + new features) 39
![Page 42: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/42.jpg)
Feature-based Approaches Spectral Feature Alignment (SFA)
Intuition Use a bipartite graph to model the correlations
between pivot features and other features Discover new shared features by applying
spectral clustering techniques on the graph
40
![Page 43: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/43.jpg)
If two domain-specific words have connections to more common pivot words in the graph, they tend to be aligned or clustered together with a higher probability. If two pivot words have connections to more common domain-specific words in the graph, they tend to be aligned together with a higher probability.
Spectral Feature Alignment (SFA) High level idea
41
exciting
good
never_buy sharp
boring
blurry
hooked
compact
realistic Pivot features
Domain-specific features
7 6
8 3
6
2
4
5
Electronics
Video Game
![Page 44: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/44.jpg)
exciting
good
never_buy sharp
boring
blurry
hooked
compact
realistic Pivot features
Domain-specific features
7 6
8 3
6
2
4
5
Electronics
Video Game
boring realistic
hooked
blurry
sharp
compact
Electronics
Video Game Electronics
Electronics Video Game
Video Game
Derive new features
42
Spectral Clustering
![Page 45: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/45.jpg)
Spectral Feature Alignment (SFA) Derive new features (cont.)
sharp/hooked compact/realistic blurry/boring 1 1 0 1 0 0 0 0 1
43
( ) sgn( ), [1,1, 1]Ty f x w x w= = ⋅ = −
sharp/hooked compact/realistic blurry/boring 1 0 0 1 1 0 0 0 1
Electronics
Video Game
Training
Prediction
![Page 46: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/46.jpg)
Spectral Feature Alignment (SFA) Algorithm
Identify P pivot features Construct a bipartite graph between the pivot and
remaining features. Apply spectral clustering on the graph to derive
new features Train classifiers on the source using augmented
features (original features + new features)
44
![Page 47: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/47.jpg)
Feature-based Approaches Develop general approaches
45
Time Period A Time Period B
Device B
Device A
![Page 48: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/48.jpg)
Feature-based Approaches General approaches
Learning features by minimizing distance between distributions Learning features inspired by multi-task
learning Learning features inspired by self-taught
learning
46
![Page 49: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/49.jpg)
Feature-based Approaches Transfer Component Analysis [Pan etal., IJCAI-09, TNN-11]
47
Target Source
Latent factors
Temperature Signal properties
Building structure
Power of APs
Motivation
![Page 50: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/50.jpg)
Transfer Component Analysis (cont.)
48
Target Source
Latent factors
Temperature Signal properties
Building structure
Power of APs
Cause the data distributions between domains different
![Page 51: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/51.jpg)
Transfer Component Analysis (cont.)
49
Target Source
Signal properties
Noisy component
Building structure
Principal components
![Page 52: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/52.jpg)
Transfer Component Analysis (cont.) Learning by only minimizing distance between distributions may map the data onto noisy factors.
50
![Page 53: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/53.jpg)
Transfer Component Analysis (cont.) Main idea: the learned should map the source and target domain data to the latent space spanned by the factors which can reduce domain difference and preserve original data structure.
51
High level optimization problem
![Page 54: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/54.jpg)
Transfer Component Analysis (cont.)
52
Recall: Maximum Mean Discrepancy (MMD)
![Page 55: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/55.jpg)
Transfer Component Analysis (cont.)
53
![Page 56: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/56.jpg)
Transfer Component Analysis (cont.)
54
The kernel function can be a highly nonlinear function of A direct optimization of minimizing the quantity w.r.t. can get stuck in poor local minima
![Page 57: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/57.jpg)
Transfer Component Analysis (cont.)
55
To maximize the data variance
To minimize the distance between domains
To preserve the local geometric structure
It is a SDP problem, expensive! It is transductive, cannot generalize on unseen instances! PCA is post-processed on the learned kernel matrix, which may
potentially discard useful information.
[Pan etal., AAAI-08]
![Page 58: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/58.jpg)
Transfer Component Analysis (cont.)
56
Parametric kernel
Minimize distance between domains
Regularization term
Maximize data variance
![Page 59: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/59.jpg)
Transfer Component Analysis (cont.)
57
An illustrative example Latent features learned by PCA and TCA
PCA Original feature space TCA
![Page 60: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/60.jpg)
Feature-based Approaches Multi-task Feature Learning
Assumption: If tasks are related, they should share some good common features.
Goal: Learn a low-dimensional representation shared across related tasks.
58
General Multi-task Learning Setting
![Page 61: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/61.jpg)
Feature-based Approaches Multi-task Feature Learning (cont.)
59
[Argyriou etal., NIPS-07]
[Ando and Zhang, JMLR-05]
[Ji etal, KDD-08]
![Page 62: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/62.jpg)
Feature-based Approaches Self-taught Feature Learning
Intuition: There exist some higher-level features that can help the target learning task even only a few labeled data are given.
Steps: 1) Learn higher-level features from a lot of unlabeled data. 2) Use the learned higher-level features to represent the data
of the target task. 3) Training models from the new representations of the
target task with corresponding labels.
60
![Page 63: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/63.jpg)
Feature-based Approaches Self-taught Feature Learning
How to learn higher-level features Sparse Coding [Raina etal., 2007] Deep learning [Glorot etal., 2011]
61
![Page 64: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/64.jpg)
Parameter-based Transfer Learning Approaches
62
Tasks are learned independently
Motivation: A well-trained model has learned a lot of structure. If two tasks are related, this structure can be transferred to learn .
![Page 65: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/65.jpg)
Parameter-based Approaches Multi-task Parameter Learning
Assumption: If tasks are related, they may share similar parameter vectors. For example, [Evgeniou and Pontil, KDD-04]
63
Common part
Specific part for individual task
![Page 66: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/66.jpg)
Parameter-based Approaches Multi-task Parameter Learning (cont.)
A general framework:
64
[Zhang and Yeung, UAI-10] [Agarwal etal, NIPS-10]
![Page 67: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/67.jpg)
Relational Transfer Learning Approaches
Motivation: If two relational domains (data is non-i.i.d) are related, they may share some similar relations among objects. These relations can be used for knowledge transfer across domains.
65
![Page 68: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/68.jpg)
Relational Transfer Learning Approaches (cont.)
66
Actor(A) Director(B) WorkedFor
Movie (M)
Student (B) Professor (A) AdvisedBy
Paper (T)
Publication Publication
Academic domain (source) Movie domain (target)
MovieMember MovieMember
AdvisedBy (B, A) ˄ Publication (B, T) => Publication (A, T)
WorkedFor (A, B) ˄ MovieMember (A, M) => MovieMember (B, M)
P1(x, y) ˄ P2 (x, z) => P2 (y, z)
[Mihalkova etal., AAAI-07, Davis and Domingos, ICML-09]
![Page 69: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/69.jpg)
camera, product, screen, photo, size, weight, quality, price, memory, etc.
great, amazing, light recommend, excellent, etc. artifacts, noise, never but, boring, etc.
Relational Approaches Relational Adaptive bootstraPping [Li etal., ACL-12]
67
Sentiment lexicon (camera)
Task: sentiment summarization What is the opinion expressed on? To construct lexicon of topic or target words
How is the opinion expressed? To construct lexicon of sentiment words
Topic lexicon (camera)
![Page 70: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/70.jpg)
This movie has good script, great casting, excellent acting. This movie is so boring. The Godfather was the most amazing movie. The movie is excellent.
The camera is great. It is a very amazing product. I highly recommend this camera. Photos had some artifacts and noise.
Relational Approaches Relational Adaptive bootstraPping (RAP) (cont.)
68
Reviews on cameras
Reviews on movies
![Page 71: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/71.jpg)
Relational Approaches RAP (cont.)
Bridge between cross-domain sentiment words – Domain independent (general) sentiment words
Bridge between cross-domain topic words
69
![Page 72: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/72.jpg)
Relational Approaches RAP (cont.)
Bridge between cross-domain topic words – Syntactic structure between topic and sentiment
words
70
Sentiment words
Topic word Topic word
Common syntactic pattern: “topic word” – nsubj – “sentiment word”
![Page 73: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/73.jpg)
Summary
71
Transfer Learning
Heterogeneous Transfer Learning
Homogeneous Transfer Learning
Unsupervised Transfer Learning
Semi-Supervised Transfer Learning
Supervised Transfer Learning
Instance-based Approaches
Feature-based Approaches
Parameter-based Approaches
Relational Approaches
In data level
In model level
![Page 74: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/74.jpg)
Some Advanced Research Issues in Transfer Learning
How to transfer knowledge across heterogeneous feature spaces
Active learning meets transfer learning
Transfer learning from multiple sources
72
![Page 75: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/75.jpg)
Reference [Thorndike and Woodworth, The Influence of Improvement in one
mental function upon the efficiency of the other functions, 1901] [Taylor and Stone, Transfer Learning for Reinforcement Learning
Domains: A Survey, JMLR 2009] [Pan and Yang, A Survey on Transfer Learning, IEEE TKDE 2009] [Quionero-Candela, etal, Data Shift in Machine Learning, MIT Press
2009] [Biltzer etal.. Domain Adaptation with Structural Correspondence
Learning, EMNLP 2006] [Pan etal., Cross-Domain Sentiment Classification via Spectral Feature
Alignment, WWW 2010] [Pan etal., Transfer Learning via Dimensionality Reduction, AAAI
2008] 73
![Page 76: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/76.jpg)
Reference (cont.) [Pan etal., Domain Adaptation via Transfer Component Analysis,
IJCAI 2009] [Evgeniou and Pontil, Regularized Multi-Task Learning, KDD 2004] [Zhang and Yeung, A Convex Formulation for Learning Task
Relationships in Multi-Task Learning, UAI 2010] [Agarwal etal, Learning Multiple Tasks using Manifold Regularization,
NIPS 2010] [Argyriou etal., Multi-Task Feature Learning, NIPS 2007] [Ando and Zhang, A Framework for Learning Predictive Structures
from Multiple Tasks and Unlabeled Data, JMLR 2005] [Ji etal, Extracting Shared Subspace for Multi-label Classification,
KDD 2008] 74
![Page 77: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/77.jpg)
Reference (cont.) [Raina etal., Self-taught Learning: Transfer Learning from Unlabeled
Data, ICML 2007] [Dai etal., Boosting for Transfer Learning, ICML 2007] [Glorot etal., Domain Adaptation for Large-Scale Sentiment
Classification: A Deep Learning Approach, ICML 2011] [Davis and Domingos, Deep Transfer vis Second-order Markov Logic,
ICML 2009] [Mihalkova etal., Mapping and Revising Markov Logic Networks for
Transfer Learning, AAAI 2007] [Li etal., Cross-Domain Co-Extraction of Sentiment and Topic
Lexicons, ACL 2012] 75
![Page 78: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/78.jpg)
Reference (cont.) [Sugiyama etal., Direct Importance Estimation with Model Selection
and Its Application to Covariate Shift Adaptation, NIPS 2007] [Kanamori etal., A Least-squares Approach to Direct Importance
Estimation, JMLR 2009] [Cristianini etal., On Kernel Target Alignment, NIPS 2002] [Huang etal., Correcting Sample Selection Bias by Unlabeled Data,
NIPS 2006] [Zadrozny, Learning and Evaluating Classifiers under Sample
Selection Bias, ICML 2004]
76
![Page 79: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/79.jpg)
Thank You
77
![Page 80: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/80.jpg)
Selected Applications of Transfer Learning
Qiang Yang and Sinno J. Pan 2013 PAKDD Tutorial Brisbane, Australia
![Page 81: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/81.jpg)
Part I. Cross Domain Transfer Learning for Activity Recognition
• Vincent W. Zheng, Derek H. Hu and Qiang Yang. Cross-Domain Activity Recognition. In Proceedings of the 11th International Conference on Ubiquitous Computing (Ubicomp-09), Orlando, Florida, USA, Sept.30-Oct.3, 2009.
• Derek Hao Hu, Qiang Yang. Transfer Learning for Activity Recognition via Sensor Mapping. In Proceedings of the 22nd International Joint Conference on Artificial Intelligence (IJCAI-11), Barcelona, Spain, July 2011
![Page 82: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/82.jpg)
3
Demo
• Annotation
![Page 83: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/83.jpg)
4
eHealth Demo
4
Sensor data
![Page 84: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/84.jpg)
5
eHealth demo
5
Activity annotation
![Page 85: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/85.jpg)
6
eHealth demo
6
Auto logging / activity recognition (service in background)
![Page 86: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/86.jpg)
Demo
• Recognition
7
![Page 87: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/87.jpg)
8
eHealth demo
8
Real-time activity recognition
![Page 88: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/88.jpg)
Demo
• Profiling
9
![Page 89: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/89.jpg)
10
eHealth demo
10
Activity profiling
![Page 90: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/90.jpg)
11
eHealth demo
11
Activity profiling for health management
![Page 91: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/91.jpg)
12
Key Problem: Recognizing Actions and Context (Locations)
Walking?
Inferred through AR
GPS and Other Sensors Sensors Sensors
Sightseeing
Watch show
Buying Ticket? Open Door?
AR: Activity Recognition via Sensors
![Page 92: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/92.jpg)
1. Cross-Domain Activity Recognition [Zheng, Hu, Yang: UbiComp-2009, PCM-2011]
• Challenge: – Some activities without data (partially labeled)
• Cross-domain activity recognition – Use other activities with available labeled data
13 Making coffee Making tea
• Happen in kitchen • Use cup, pot • …
![Page 93: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/93.jpg)
14
Cleaning Indoor
Laundry
Dishwashing
![Page 94: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/94.jpg)
15
System Workflow
15
Source Domain Labeled Data
Similarity Measure
<Sensor Reading, Activity Name>
Example: <SS, “Make Coffee”>
Example: sim(“Make Coffee”, “Make Tea”) = 0.6
Target Domain Pseudo Labeled
Data
Weighted SVM Classifier
THE WEB
Example: Pseudo Training Data: <SS, “Make Tea”, 0.6>
![Page 95: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/95.jpg)
Calculating Activity Similarities
How similar are two activities? ◦ Use Web search
results ◦ TFIDF: Traditional IR
similarity metrics (cosine similarity) ◦ Example Mined similarity between
the activity “sweeping” and “vacuuming”, “making the bed”, “gardening”
Calculated Similarity with the activity
"Sweeping"
16 16
![Page 96: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/96.jpg)
17
Datasets: MIT PlaceLab http://architecture.mit.edu/house_n/placelab.html
• MIT PlaceLab Dataset (PLIA2) [Intille et al. Pervasive 2005]
• Activities: Common household activities
17
![Page 97: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/97.jpg)
18
Datasets: Intel Research Lab
• Intel Research Lab [Patterson, Fox, Kautz, Philipose, ISWC2005] – Activities Performed:
11 activities – Sensors
• RFID Readers & Tags
– Length: • 10 mornings
18 Picture excerpted from [Patterson, Fox, Kautz, Philipose, ISWC2005].
![Page 98: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/98.jpg)
Cross-Domain AR: Performance
Accuracy with Cross Domain Transfer
# Activities (Source Domain)
# Activities (Target Domain)
Baseline (Random Guess)
Supervised (Upper bound)
Intel Research Lab Dataset
63.2% 5 6 16.7% 78.3%
Amsterdam Dataset 65.8% 4 3 33.3% 72.3% MIT Dataset (Cleaning to Laundry)
58.9% 13 8 12.5% -
MIT Dataset (Cleaning to Dishwashing)
53.2% 13 7 14.3% -
19 19
Activities in the source domain and the target domain are generated from ten random trials, mean accuracies are reported.
![Page 99: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/99.jpg)
Derek Hao Hu and Qiang Yang, IJCAI 2011
Transferring Across
Feature Space
Transferring Across
Label Space
Transfer from Source Domain to
Target Domain ( )
( | ) ( | ) ( | )i
s
t t t tc
p y x p c x p y c∈
= ⋅∑L
![Page 100: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/100.jpg)
Proposed Approach • Final goal: Estimate
– We have
– Estimating the above equation at its mode:
)( |t tp y x
Feature Transfer Label Transfer
![Page 101: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/101.jpg)
Experiments • Datasets
– UvA dataset [van Kasteren et al. Ubicomp 2008] – MIT Placelab (PLIA1) dataset [Intille et al. Ubicomp 2006] – Intel Research Lab dataset [Patterson et al. ISWC 2005]
• Baseline – Unsupervised Activity Recognition Algorithm [Wyatt et al. 2005]
• Different sensors for different datasets
State-based sensors for UvA dataset A series of different wired
sensors for MIT dataset
RFID sensor for Intel Research Lab
Dataset
![Page 102: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/102.jpg)
Experiments: Different Feature & Label Spaces
• Source: MIT PLIA1 dataset Target: UvA (Intel) datasets
![Page 103: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/103.jpg)
Part II
• Source Free Transfer Learning • Evan Wei Xiang, Sinno Jialin Pan, Weike Pan, Jian Su and Qiang
Yang. Source-Selection-Free Transfer Learning. In Proceedings of the 22nd International Joint Conference on Artificial Intelligence (IJCAI-11), Barcelona, Spain, July 2011.
![Page 104: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/104.jpg)
Source-Selection-Free Transfer Learning
Evan Xiang, Sinno Pan, Weike Pan, Jian Su, Qiang Yang
HKUST - IJCAI 2011 25
![Page 105: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/105.jpg)
Transfer Learning
Lack of labeled training data
always happens
When we have some related
source domains
Supervised Learning
Transfer Learning
HKUST - IJCAI 2011 26
![Page 106: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/106.jpg)
Where are the “right” source data?
HKUST - IJCAI 2011 27
We may have an extremely large number of choices of potential sources to use.
![Page 107: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/107.jpg)
Outline of Source-Selection-Free Transfer Learning (SSFTL)
Stage 1: Building base models
Stage 2: Label Bridging via Laplacian Graph Embedding
Stage 3: Mapping the target instance using the base classifiers & the projection matrix
Stage 4: Learning a matrix W to directly project the target instance to the latent space
Stage 5: Making predictions for the incoming test data using W
HKUST - IJCAI 2011 28
![Page 108: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/108.jpg)
SSFTL – Building base models
vs.
vs.
vs.
vs.
vs.
vs.
vs.
vs.
vs.
vs.
vs.
From the taxonomy of the online information source, we can “Compile” a lot of base classification models
HKUST - IJCAI 2011 29
![Page 109: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/109.jpg)
SSFTL – Label Bridging via Laplacian Graph Embedding
vs.
vs.
vs.
vs.
vs.
vs.
mism
atch
However, the label spaces of the based classification
models and the target task can be different
The relationships between labels, e.g., similar or dissimilar, can be represented by the distance between their corresponding prototypes in
the latent space, e.g., close to or far away from each other.
Since the label names are usually short and sparse, , in order to uncover the intrinsic
relationships between the target and source
labels, we turn to some social media such as Delicious, which can help to
bridge different label sets together.
V
Projection matrix
q
m
HKUST - IJCAI 2011 30
Bob
Tom
John
Gary
Steve Sports
Tech
Finance
Travel
History
M q
q
Neighborhood matrix for label graph
Problem
m-dimensional latent space
Laplacian Eigenmap [Belkin & Niyogi,2003]
![Page 110: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/110.jpg)
SSFTL – Mapping the target instance using the base classifiers & the projection matrix V vs.
vs.
vs.
vs.
vs.
For each target instance, we can obtain a combined result on the label space via aggregating the predictions
from all the base classifiers
However, do we need to recall the base classifiers during the prediction phase? The answer is No!
Then we can use the projection matrix V to transform such combined results from
the label space to a latent space
V
Projection matrix
q
m
Prob
abili
ty
Label space
“Ipad2 is released in March, …”
HKUST - IJCAI 2011 31
Sports
Tech
Finance
Travel
History
Target Instance 0.1:0.9
0.3:0.7
0.2:0.8
0.6:0.4
0.7:0.3
q
= <Z1, Z2, Z3, …, Zm>
m-dimensional latent space
![Page 111: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/111.jpg)
SSFTL – Learning a matrix W to directly project the target instance to the latent space
vs.
vs.
vs.
vs.
vs. V
Projection matrix Target Domain
Labeled & Unlabeled
Data
q
m
W d
m
Learned Projection matrix
Our regression model
Loss on labeled data
Loss on unlabeled data
For each target instance, we first aggregate its prediction on the base label space, and
then project it onto the latent space
HKUST - IJCAI 2011 32
![Page 112: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/112.jpg)
SSFTL – Making predictions for the incoming test data
vs.
vs.
vs.
vs.
vs. V
Projection matrix
Target Domain
Incoming Test Data
q
m
W d
m
Learned Projection matrix
Therefore, we can make prediction directly for any
incoming test data based on the distance to the label prototypes,
without calling the base classification models
The learned projection matrix W can be used to transform any target instance directly
from the feature space to the latent space
HKUST - IJCAI 2011 33
![Page 113: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/113.jpg)
Experiments - Datasets Building Source Classifiers with Wikipedia 3M articles, 500K categories (mirror of Aug 2009) 50, 000 pairs of categories are sampled for source models
Building Label Graph with Delicious 800-day historical tagging log (Jan 2005 ~ March 2007) 50M tagging logs of 200K tags on 5M Web pages
Benchmark Target Tasks 20 Newsgroups (190 tasks) Google Snippets (28 tasks) AOL Web queries (126 tasks) AG Reuters corpus (10 tasks)
HKUST - IJCAI 2011 34
![Page 114: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/114.jpg)
SSFTL - Building base classifiers Parallelly using MapReduce
Input Map Reduce
The training data are replicated and assigned to different bins
In each bin, the training data are paired for building binary
base classifiers
vs.
vs. vs.
vs. 1
2
3
1 3 …
2 3 …
1 2 …
…
…
…
If we need to build 50,000 base classifiers, it would take about two days if we run the training process on a single server. Therefore, we distributed the training process to a cluster with 30 cores using MapReduce, and finished the training within two hours.
These pre-trained source base classifiers are stored and reused for different incoming target tasks.
HKUST - IJCAI 2011 35
![Page 115: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/115.jpg)
Experiments - Results
HKUST - IJCAI 2011 36
-Parameter setttings- Source models: 5,000 Unlabeled target data: 100% lambda_2: 0.01
Semi-supervised SSFTL Unsupervised SSFTL
Our regression model
![Page 116: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/116.jpg)
Experiments - Results
HKUST - IJCAI 2011 37
-Parameter setttings- Mode: Semi-supervised Labeled target data: 20 Unlabeled target data: 100% lambda_2: 0.01
Our regression model
Loss on unlabeled data
For each target instance, we first aggregate its prediction on the base label space, and
then project it onto the latent space
![Page 117: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/117.jpg)
Experiments - Results
HKUST - IJCAI 2011 38
-Parameter setttings- Mode: Semi-supervised Labeled target data: 20 Source models: 5,000 lambda_2: 0.01
Our regression model
![Page 118: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/118.jpg)
Experiments - Results
HKUST - IJCAI 2011 39
-Parameter setttings- Labeled target data: 20 Unlabeled target data: 100% Source models: 5,000
Semi-supervised SSFTL Supervised SSFTL
Our regression model
![Page 119: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/119.jpg)
Experiments - Results
HKUST - IJCAI 2011 40
-Parameter setttings- Mode: Semi-supervised Labeled target data: 20 Source models: 5,000 Unlabeled target data: 100% lambda_2: 0.01 Our regression model
Loss on unlabeled data
For each target instance, we first aggregate its prediction on the base label space, and
then project it onto the latent space
![Page 120: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/120.jpg)
Related Works
HKUST - IJCAI 2011 41
![Page 121: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/121.jpg)
Conclusion Source-Selection-Free Transfer Learning When the potential auxiliary data is embedded in very
large online information sources
No need for task-specific source-domain data We compile the label sets into a graph Laplacian for
automatic label bridging
SSFTL is highly scalable Processing of the online information source can be done
offline and reused for different tasks.
HKUST - IJCAI 2011 42
![Page 122: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/122.jpg)
Q & A
HKUST - IJCAI 2011 43
![Page 123: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/123.jpg)
Advance Research Topics in Transfer Learning
Wei Fan
Huawei Noah's Ark Research Lab, Hong Kong
![Page 124: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/124.jpg)
Predictive Modeling with Heterogeneous Sources
Xiaoxiao Shi Qi Liu Wei Fan
Qiang Yang Philip S. Yu
![Page 125: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/125.jpg)
1/18
Why learning with heterogeneous sources?
New York Times
Training (labeled)
Test (unlabeled)
Classifier
New York Times
85.5%
Standard Supervised Learning
![Page 126: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/126.jpg)
2/18 New York Times
Training (labeled)
Test (unlabeled)
New York Times Labeled data are
insufficient! 47.3%
How to improve the performance?
In Reality…
Why heterogeneous sources?
![Page 127: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/127.jpg)
3/18
Why heterogeneous sources?
Reuters
Labeled data from other sources
Target domain test (unlabeled)
New York Times
82.6%
1. Different distributions
2. Different outputs
3. Different feature spaces
47.3%
![Page 128: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/128.jpg)
Real world examples
• Social Network: – Can various bookmarking systems help predict social tags for a
new system given that their outputs (social tags) and data (documents) are different?
Wikipedia ODP Backflip Blink ……
? 4/18
![Page 129: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/129.jpg)
Real world examples
• Applied Sociology: – Can the suburban housing price census data help predict the
downtown housing prices?
?
#rooms #bathrooms #windows price
5 2 12 XXX
6 3 11 XXX
#rooms #bathrooms #windows price
2 1 4 XXXXX
4 2 5 XXXXX 5/18
![Page 130: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/130.jpg)
Other examples
• Bioinformatics – Previous years’ flu data new swine flu – Drug efficacy data against breast cancer
drug data against lung cancer – ……
• Intrusion detection – Existing types of intrusions unknown
types of intrusions • Sentiment analysis
– Review from SDM Review from KDD
6/18
![Page 131: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/131.jpg)
Learning with Heterogeneous Sources
• The paper mainly attacks two sub-problems: – Heterogeneous data distributions
• Clustering based KL divergence and a corresponding sampling technique
– Heterogeneous outputs (to regression problem)
• Unifying outputs via preserving similarity.
7/18
![Page 132: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/132.jpg)
Learning with Heterogeneous Sources
• General Framework
Unifying data distributions Unifying outputs
Source data
Target data
Source data Target data
8/18
![Page 133: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/133.jpg)
Unifying Data Distributions
• Basic idea: – Combine the source and target data and
perform clustering. – Select the clusters in which the target and
source data are similarly distributed, evaluated by KL divergence.
9/18
![Page 134: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/134.jpg)
An Example D T
Combined Data
Adaptive Clustering
10/18
![Page 135: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/135.jpg)
Unifying Outputs
• Basic idea: – Generate initial outputs according to the
regression model – For the instances similar in the original output
space, make their new outputs closer.
11/18
![Page 136: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/136.jpg)
12/18
16 37 26.5 21.25 31.75
Initial Outputs
Initial Outputs
![Page 137: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/137.jpg)
Experiment
• Bioinformatics data set:
13/18
![Page 138: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/138.jpg)
Experiment
14/18
![Page 139: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/139.jpg)
Experiment
• Applied sociology data set:
15/18
![Page 140: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/140.jpg)
Experiment
16/18
![Page 141: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/141.jpg)
17/18
• Problem: Learning with Heterogeneous Sources: • Heterogeneous data distributions • Heterogeneous outputs
• Solution: • Clustering based KL divergence help perform
sampling • Similarity preserving output generation help
unify outputs
Conclusions
![Page 142: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/142.jpg)
Transfer Learning on Heterogeneous Feature Spaces via Spectral Transformatio
Xiaoxiao Shi, Qi Liu, Wei Fan, Philip S. Yu, and Ruixin Zhu
![Page 143: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/143.jpg)
Motivation
1/18
Training documents (labeled)
Test documents (unlabeled)
85.5%
Standard Supervised Learning
Classifier
![Page 144: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/144.jpg)
Motivation
Training (labeled)
Huge set of unlabeled documents
Labeled data are insufficient!
47.3%
How to improve the performance?
In Reality…
![Page 145: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/145.jpg)
Learning Formulations
![Page 146: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/146.jpg)
Learning from heterogeneous sources
3/18
Labeled data from other sources
Target domain test (unlabeled)
???
Heterogeneous datasets: 1.Different data distributions: P(xtrain) and P(xtest) are different
2.Different outputs: ytrain and ytest are different 3.Different feature spaces: xtrain and xtest are different
![Page 147: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/147.jpg)
Some Applications of Transfer Learning
• WiFi-based localization tracking [Pan et al'08]
• Collaborative Filtering [Pan et al'10] • Activity Recognition [Zheng et al'09] • Text Classification [Dai et al'07] • Sentiment Classification [Blitzer et al‘07] • Image Categorization [Shi et al’10] • … …
![Page 148: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/148.jpg)
Issues • Different data distributions: P(xtrain) and P(xtest)
are different
focuses more on Chicago local news
focuses more on global news
focuses more on scientific/objective documents
![Page 149: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/149.jpg)
Issues • Different outputs: ytrain and ytest are
different Wikipedia ODP Yahoo!
![Page 150: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/150.jpg)
Issues
• Different feature spaces (the focus on the paper) – Drug efficacy tests:
• Physical properties • Topological properties
– Image Classification • Wavelet features • Color histogram
![Page 151: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/151.jpg)
Unify different feature spaces
• Different number of features; different meanings of the features, no common feature, no overlap.
• Projection-based approach HeMap – Find a projected space where (1) the source and
target data are similar in distribution; (2) the original structure (separation) of each of the dataset is preserved.
![Page 152: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/152.jpg)
Unify different feature spaces via HeMap
Optimization objective of HeMap:
The linear projection error
The linear projection error
The difference between the projected data
![Page 153: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/153.jpg)
Unify different feature spaces via HeMap
With some derivations, the objective can be reformulated as (more details can be found in the paper):
![Page 154: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/154.jpg)
Algorithm flow of HeMap
![Page 155: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/155.jpg)
Generalized HeMap to handle heterogeneous data (different distributions, outputs and feature spaces)
![Page 156: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/156.jpg)
Unify different distributions and outputs
• Unify different distributions – Clustering based sample selection [Shi etc
al,09] • Unify different outputs
– Bayesian like schema
![Page 157: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/157.jpg)
Generalization bound
Principle II: minimize the combined expected error by maintaining the original
structure (minimize projection error)
Principle I: minimize the difference between target
and source datasets
and are domain-specific
parameters; is model
complexity
![Page 158: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/158.jpg)
Experiments
• Drug efficacy prediction – The dataset is collected by the College of Life
Science and Biotechnology of Tongji University, China. It is to predict the efficacy of drug compounds against certain cell lines.
– The data are generated in two different feature spaces
• general descriptors: refer to physical properties of compounds
• drug-like index: refer to simple topological indices of compounds.
![Page 159: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/159.jpg)
Experiments
![Page 160: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/160.jpg)
Experiments
• Image classification
Cartman &
Bonsai
Homer Sim
pson &
Cactus
Homer Sim
pson &
Coin
Superman &
CD
![Page 161: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/161.jpg)
Experiments
![Page 162: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/162.jpg)
Conclusions • Extends the applicability of supervised
learning, semi-supervised learning and transfer learning by using heterogeneous data: – Different data distributions – Different outputs – Different feature spaces
• Unify different feature spaces via linear projection with two principles – Maintain the original structure of the data – Maximize the similarity of the two data in the
projected space
![Page 163: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/163.jpg)
Cross Validation Framework to Choose Amongst Models and Datasets for Transfer Learning
Erheng Zhong¶, Wei Fan‡, Qiang Yang¶, Olivier Verscheure‡, Jiangtao Ren†
![Page 164: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/164.jpg)
Transfer Learning: What is it
Applications
Definition
“source-domains” to improve “target-domain”: short of labeled information.
• supervised • unsupervised
• semi-supervised • transfer learning
1. WiFi-based localization tracking [Pan et al'08] 2. Collaborative Filtering [Pan et al'10] 3. Activity Recognition [Zheng et al'09]
4. Text Classification [Dai et al'07] 5. Sentiment Classification [Blitzer et al‘07]
6. Image Categorization [Shi et al’10] .......
![Page 165: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/165.jpg)
Application Indoor WiFi localization tracking
AP is the access point of device. (Lx, Ly) is the coordinate of location.
Transfer
![Page 166: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/166.jpg)
Application Collaborative Filtering
![Page 167: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/167.jpg)
Transfer Learning: How it works Data Selection
Model Selection
![Page 168: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/168.jpg)
Re-cast: Model and Data Selection
(1) How to select the right transfer learning algorithms?
(2) How to tune the optimal parameters?
(3) How to choose the most helpful source-domain from a large pool of datasets?
![Page 169: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/169.jpg)
Model & Data Selection Traditional Methods
1. Analytical techniques: AIC, BIC, SRM, etc.
2. k-fold cross validation
![Page 170: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/170.jpg)
Model & Data Selection Issuses
)()( xPxP ts ≠
The estimation is not consistent.
)|()|( xyPxyP ts ≠
A model approximating is not necessarily close to
The number of labeled data in target domain is limited and thus the directly estimation of is not reliable.
)|( xyPs )|( xyPt
)|( xyPt
Ideal Hypothesis
![Page 171: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/171.jpg)
Model & Data Selection Model Selection Example
Source
Target
If we choose the wrong model....
![Page 172: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/172.jpg)
Model & Data Selection Data Selection Example
Target
If we choose the wrong source-domain....
![Page 173: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/173.jpg)
Transfer Cross-Validation (TrCV) New criterion for transfer learning
Hard to calculate
in practice
1. The density ration between two domains
2. The difference between the conditional distribution estimated by model and the true conditional distribution.
Reverse Validation
Practical method: Transfer Cross-Validation (TrCV)
How to calculate this difference with
limited labeled data?
Density Ratio Weighting
![Page 174: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/174.jpg)
Density Ratio Weighting • The selected model is an unbiased estimator to the ideal
model
is the expected loss to approximate
is the model complexity
• We adopt an existing method KMM (Huang et al’07) for density ratio weighting
• Reverse Validation to estimate Pt(y|x) – P(y|x,f) (next slide)
Important property to choose the right model even when P(x) and P(y|x) are different
![Page 175: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/175.jpg)
Reverse Validation
The source-domain data in i-th fold
The remaining data The predicted label of in i-th fold
The predicted label of in i-th fold
The true label of in i-th fold
The unlabeled and labeled target-domain data
![Page 176: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/176.jpg)
Properties
• The model selected by the proposed method has a generalization bound over target-domain data. [Theorem 1] • The value of reverse validation is related to the
difference between true conditional probability and model approximation.
• The confidence of TrCV has a bound.
the accuracy estimated by TrCV the true accuracy of quantile point of the standard normal distribution
• The selected model is an unbiased estimator to the ideal one. [Lemma 1]
![Page 177: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/177.jpg)
Experiment Data Set
• Wine Quality: two subsets related to red and white variants of the Portuguese “Vinho Verde” wine.
For algorithm and parameters selection
![Page 178: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/178.jpg)
Experiment Data Set
• Reuters-21578:the primary benchmark of text categorization formed by different news with a hierarchial structure.
For algorithm and parameters selection
![Page 179: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/179.jpg)
Experiment Data Set
• SyskillWebert: the standard dataset used to test web page ratings, generated by the HTML source of web pages plus the user rating. we randomly reserve “Bands-recording artists” as source-domain and the three others as target-domain data.
For algorithm and parameters selection
![Page 180: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/180.jpg)
Experiment Data Set
• 20-Newsgroup: primary benchmark of text categorization similar to Reuters-21578
For source-domain selection
![Page 181: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/181.jpg)
Experiment Baseline methods
• SCV: standard k-fold CV on source-domain • TCV: standard k-fold CV on labeled data from target-
domain • STV: building a model on the source-domain data and
validating it on labeled target-domain data • WCV: using density ratio weighting to reduce the
difference of marginal distribution between two domains, but ignoring the difference in conditional probability.
![Page 182: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/182.jpg)
Experiment Other settings
• Algorithms: – Naive Bayes(NB), SVM, C4.5, K-NN and NNge(Ng) – TrAdaBoost(TA): instances weighting [Dai et al.'07] – LatentMap(LM): feature transform [Xie et al.'09] – LWE : model weighting ensemble [Gao et al.'08]
• Evaluation: if one criterion can select the better model in the comparison, it gains a higher measure value.
The accuracy and value of criteria (e.g TrCV, SCV, etc)
The number of comparisions between models
![Page 183: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/183.jpg)
Results Algorithm Selection
6 win and 2 lose!
![Page 184: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/184.jpg)
Results Parameter Tuning
13 win and 3 lose!
![Page 185: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/185.jpg)
Results Source-domain Selection
No lose!
![Page 186: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/186.jpg)
Results Parameter Analysis
TrCV achieves the highest correlation value under different number of folds from 5 to 30 with step size 5 .
![Page 187: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/187.jpg)
Results Parameter Analysis
When only a few labeled data(< 0.4 × |T|) can be obtained in the target-domain, the performance of TrCV is much better
than both SVT and TCV.
![Page 188: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/188.jpg)
Conclusion
• Model and data selection when margin and conditional distributions are different between two domains.
• Key points – Point-1 Density weighting to reduce the difference
between marginal distributions of two domains; – Point-2 Reverse validation to measure how well a
model approximates the true conditional distribution of target-domain.
• Code and data available from the authors – www.weifan.info
![Page 189: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/189.jpg)
![Page 190: Transfer Learning with Applicationsijcai13.org/files/tutorial_slides/td2.pdf · Huawei Noah's Ark Research Lab, ... Parameter-based Approaches . Relational ... Use the same strategy](https://reader035.fdocuments.net/reader035/viewer/2022062504/5abc58567f8b9a297f8df9b1/html5/thumbnails/190.jpg)
18/18
Thanks!