Language Models as Representations for Weakly-Supervised NLP Tasks (CoNLL2011)
Transfer for Supervised Learning Tasks
description
Transcript of Transfer for Supervised Learning Tasks
![Page 1: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/1.jpg)
HAITHAM BOU AMMAR
MAASTRICHT UNIVERSITY
Transfer for Supervised Learning Tasks
![Page 2: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/2.jpg)
Motivation
Model
€
y
€
x
€
x€
y
Training Data
?
Test Data
Assumptions: 1.Training and Test are from same distribution 2.Training and Test are in same feature spaceNo
t Tru
e in
man
y re
al-
world
app
licat
ions
![Page 3: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/3.jpg)
Examples: Web-document Classification
Model
?
Physics Machine Learning
Life Science
![Page 4: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/4.jpg)
Examples: Web-document Classification
Model
?
Physics Machine Learning
Life Science
Content Change !
Assumption violated!
Learn a new model
![Page 5: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/5.jpg)
Learn new Model :
1.Collect new Labeled Data2.Build new model
Reuse & Adapt already learned model !
![Page 6: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/6.jpg)
Examples: Image Classification
Model OneFeatures Task One
Task One
![Page 7: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/7.jpg)
Examples: Image Classification
Cars
Motorcycles
Task Two
Features Task One
Features Task Two
Reuse
Model Two
![Page 8: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/8.jpg)
Traditional Machine Learning vs. Transfer
Source Task
Knowledge
Target Task
Learning System
Different Tasks
Learning System
Learning System
Learning System
Traditional Machine Learning
Transfer Learning
![Page 9: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/9.jpg)
Questions ?
![Page 10: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/10.jpg)
Transfer Learning Definition
Notation: Domain :
Feature Space: Marginal Probability Distribution:
with
Given a domain then a task is :
Labe
l
Space P(Y
|X)
![Page 11: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/11.jpg)
Transfer Learning Definition
Given a source domain and source learning task, a target domain and a target learning task, transfer learning aims to help improve the learning of the target predictive function using the source knowledge, where
or
![Page 12: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/12.jpg)
Transfer Definition
Therefore, if either : Domain Differences
Task Differences
![Page 13: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/13.jpg)
Examples: Cancer Data
Age Smoking
Age Height Smoking
![Page 14: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/14.jpg)
Examples: Cancer Data
Task S
ourc
e: Clas
sify
into c
ance
r or n
o
canc
erTa
sk T
arge
t: Clas
sify
into c
ance
r lev
el on
e,
canc
er le
vel t
wo,
canc
er le
vel t
hree
![Page 15: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/15.jpg)
Quiz
When doesn’t transfer help ? (Hint: On what should you
condition?)
![Page 16: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/16.jpg)
Questions ?
![Page 17: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/17.jpg)
Questions to answer when transferringW
hat t
o Tra
nsfe
r ?
How to
Trans
fer ?
Whe
n to
Tra
nsfe
r ?
Instan
ces
?
Model
?
Features
?
Map
M
odel
?
Unify
Feat
ures
?
Weig
ht
Insta
nces
?
In w
hich
Si
tuat
ions
![Page 18: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/18.jpg)
Algorithms: TrAdaBoost
Assumptions: Source and Target task have same feature space:
Marginal distributions are different:
Not all source data might be helpful !
![Page 19: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/19.jpg)
Algorithms: TrAdaBoost (Quiz)
Wha
t to
Trans
fer ?
How to
Tra
nsfer
?
Instan
ces
Weig
ht In
stanc
es
![Page 20: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/20.jpg)
Algorithm: TrAdaBoost
Idea: Iteratively reweight source samples such that:
reduce effect of “bad” source instances encourage effect of “good” source instances
Requires: Source task labeled data set Very small Target task labeled data set Unlabeled Target data set Base Learner
![Page 21: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/21.jpg)
Algorithm: TrAdaBoost
Weights Initialization
Hypothesis Learning and error calculation
Weights Update
![Page 22: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/22.jpg)
Questions ?
![Page 23: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/23.jpg)
Algorithms: Self-Taught Learning
Problem Targeted is :
1.Little labeled data
2.A lot of unlabeled data
Build a model on the labeled data
Natural scenes
Car
Motorcycl
e
![Page 24: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/24.jpg)
Algorithms: Self-Taught Learning
Assumptions: Source and Target task have different feature space:
Marginal distributions are different:
Label Space is different:
![Page 25: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/25.jpg)
Algorithms: Self-Taught Learning (Quiz)
Wha
t to
Trans
fer ?
How to
Tra
nsfer
?
1. Disc
over
Featu
res
2. Unif
y Fea
ture
s
Features
![Page 26: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/26.jpg)
Algorithms: Self-Taught Learning
Framework: Source Unlabeled data set:
Target Labeled data set: Natural
scenes
Build classifier for cars and Motorbikes
![Page 27: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/27.jpg)
Algorithms: Self-Taught Learning
Step One: Discover high level features from Source data by
Regularization Term Re-construction Error
Constraint on the Bases
![Page 28: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/28.jpg)
Algorithm: Self-Taught Learning
Unlabeled Data Set
High
Leve
l Fea
ture
s
![Page 29: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/29.jpg)
Algorithm: Self-Taught Learning
Step Two: Project target data onto the attained features by
Informally, find the activations in the attained bases such that: 1.Re-construction is minimized 2.Attained vector is sparse
![Page 30: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/30.jpg)
Algorithms: Self-Taught Learning
High Level F
eatures
![Page 31: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/31.jpg)
Algorithms: Self-Taught Learning
Step Three: Learn a Classifier with the new features
Target TaskSource Task
Learn new features (Step 1)
Project target data (Step 2)Learn Model (Step 3)
![Page 32: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/32.jpg)
Conclusions
Transfer learning is to re-use source knowledge to help a target learner
Transfer learning is not generalization
TrAdaBoot transfers instances
Self-Taught learning transfer unlabeled features
![Page 33: Transfer for Supervised Learning Tasks](https://reader036.fdocuments.us/reader036/viewer/2022062810/56815dfe550346895dcc3e59/html5/thumbnails/33.jpg)
hmm