Continuously Indexed Domain Adaptation - PowerPoint Presentation

Slide1

Continuously Indexed Domain Adaptation

(* indicates equal contribution)

Hao He*

Dina

Katabi

Hao

Wang

*

ICML 2020 Oral

Slide2

Domain Adaptation

One to One

Source Domain

Target Domain

and

 

 

Many to One

Single Target Domain

Multiple Source Domains

Many to Many

Multiple Source Domains

Multiple Target Domains

Predict

 

Slide3

MNIST

SVHN

RGB

Depth

Problem: Current Domain Adaptation is Categorical

Category 1

Category 2

But many real problem involves continuous domains!

Slide4

Medical Applications Require Adaptation Across Age Which is

Continuous

Young

Old

Age

Slide5

Self-Driving Car Applications Require Adaptation Across Time Which is

Continuous

Morning

Night

Time

Slide6

Example of

Continuously Indexed Domain Adaptation Problems

Domain 1

Domain 2

Domain 15

Domain 30

Setup:

30 continuously indexed domains

Ground-truth labels (

red

and

blue

)

Source:

Domain 1 to 6

Target:

Domain 7 to 30

Task:

Source -> Target

Slide7

Performance of Categorical DA

Source

Target

Results

ADDA

:

Adversarial Discriminative Domain Adaptation

(Tzeng, Eric, et al., CVPR 2017)

Slide8

Performance of Categorical DA

Source

Target 1

Target 2

Target 3

DANN

: Domain-Adversarial Training of Neural Networks

(

Ganin

,

Yaroslav

, et al., JMLR 2016)

Results

How should we perform

continuous domain adaptation?

Slide9

Why Categorical Domain Adaptation Fails

Source:

domain 1

Target:

domain 8

Cannot

capture how the data distribution continuously changes along with the domain index

Target:

domain 15

Distant

Close

Slide10

Why Categorial Domain Adaptation Fails

Source:

domain 1

Target:

domain 15

Target:

domain 8

Slide11

Solution: Leverage the Domain Index

Source:

domain 1

Target:

domain 15

Target:

domain 8

1

8

15

Slide12

Categorical Domain Adaptation

Data

 

 

Predicted label

 

Source or target?

Encoder E

Predictor F

Discriminator D

E

F

D

Slide13

Continuously Indexed Domain Adaptation (CIDA)

Data

 

Domain index

 

 

Predicted label

 

Predicted domain index

 

Encoder E

Predictor F

Discriminator D

E

F

D

For

:

Minimize

 

For

,

:

Minimize

 

Slide14

Challenge for CIDA

D

omain 1

D

omain 3

D

omain 2

Before training,

Slide15

Challenge for CIDA

Ideally, after training

 

 

 

 

CIDA may end up with

CIDA only matches

 

 

 

Slide16

To Improve CIDA

Ideally, after training

 

 

 

 

 

 

CIDA may end up with

 

Slide17

Continuously Indexed Domain Adaptation (CIDA)

Data

 

Domain index

 

 

Predicted label

 

Predicted domain index

 

Encoder E

Predictor F

Discriminator D

E

F

D

For

:

Minimize

 

For

,

:

Minimize

 

Slide18

Probabilistic CIDA (PCIDA)

Data

 

Domain index

 

 

Predicted label

 

Predicted mean

 

Encoder E

Predictor F

Discriminator D

For

:

Minimize

 

Predicted variance

 

E

F

D

For

,

:

Minimize

 

PCIDA matches both

and

 

Slide19

Theoretical Analysis

Theorem 1

(Informal). CIDA converges,

if and only if

, the expectation of domain index

is identical for any embedding

.

 

Theorem 2

(Informal). PCIDA converges, if and only if, the expectation and the variance of domain index,

and

are identical for any

.

 

Theorem 3

(Informal). The global optimum of the two-player game between

and

matches the global optimum of the three-play game between

,

, and

.

 

Slide20

Single-Step

Categorical Domain AdaptationADDA: Adversarial Discriminative Domain AdaptationDANN: Domain-Adversarial Training of Neural Networks… (more results in the paper)

Baselines

Multi-Step Categorical Domain Adaptation

CUA : Continuous Unsupervised Adaptation

Domain 1

Domain 2

Domain 3

Domain 4

Slide21

Experiments -

Circle

Domain 1

Domain 2

Domain 15

Domain 30

30 continuously indexed domains

Ground-truth labels (

red

and

blue

)

Slide22

Experiments -

Circle

The

first 6 domains

as source domains

with the rest as target domains

Ground-truth labels (

red

and blue)

Slide23

Experiments -

Circle

DANN

CIDA

CUA

ADDA

Ground truth

Slide24

Results (Zoomed in)

Result of applying CIDA

Source

T

arget Part 1

T

arget Part 2

Result of applying CUA

CUA: Local adaptation. Very rugged boundary.

CIDA: Global adaptation. Very

smooth boundary.

Slide25

Experiments -

Sine

Similar slides to Circle

Domain 1

Domain 2

Domain 6

Domain 12

12 continuously indexed domains

Ground-truth labels (

red

and

blue

)

Slide26

Experiments -

Sine

Similar slides to Circle

Ground-truth labels (

red

and

blue)

The

first 5 domains as source domains

with the rest as target domains

12 continuously indexed domains

Slide27

Experiments -

Sine

DANN

CIDA

CUA

ADDA

Ground truth

Slide28

Experiments -

Sine

CUA

Domain 5 and 6

Domain 6 and 7

Domain 7 and 8

Domain 8 and 9

Ground truth

CUA may fail in an intermediate domain,

causing failure in all following domains.

Slide29

Experiments -

Sine

CIDA

Ground truth

Slide30

Real-World Medical Scenario

Sleep Study at Home

Predictor

1. Awake

2. REM

3. Light Sleep

4. Deep Sleep

Input Time-Series Signals:

2. Breathing Belt

1. Nasal Cannula

Slide31

Evaluation on Real-World Datasets

SHHS

: Sleep Heart Health StudyMESA

: Multi-Ethnic Study of AtherosclerosisSOF

: Study of Osteoporotic Fractures

Three Sleep Study Datasets

Slide32

Continuous Adaptation Setting: Extrapolation

Young

Old

Age

Here ‘age’ is a

domain index

.

Source domains

Slide33

Continuous Adaptation Setting: Interpolation

Young

Old

Age

Here ‘age’ is a

domain index

.

Source domains

Source domains

Slide34

Results on Extra/Inter-

polation Adaptation

Categorical DA may hurt performance.

Source-only

Categorical DA

Ours

CIDA/PCIDA improve performance in both settings.

CIDA/PCIDA has larger performance gain in extrapolation since it is harder.

Slide35

Results on Cross Dataset Adaptation

In the

hardest

transferring tasks (

SOF -> SHHS

/MESA),

PCIDA is a clear winner.

Slide36

Multi-Dimensional Domain Index

Domain Index 1: Age

Domain Index 2:

Physical Wellness

Domain Index 3:

Emotional Wellness

Slide37

Multi-Dimensional Domain Index

Multi-dimensional indices further improve performance

Multi-dimensional domain index:

Age

Physical wellness

Emotional wellness

Fatigue level

…

Slide38

Summary

The first general DA method for continuously indexed domain adaptation.

Theoretical guarantees that CIDA aligns continuously indexed domains at equilibrium.

Two advanced versions, probabilistic CIDA (PCIDA) and

multi-dimensional CIDA.State-of-the-art performance on both synthetic and real-world medical datasets.

Code

will be released soon at https://github.com/hehaodele/CIDA.