H - - PowerPoint PPT Presentation

▶
h
SMART_READER_LITE
LIVE PREVIEW

H - - PowerPoint PPT Presentation

Aug 06, 2023 397 likes •999 views

Symposium What Neuroimaging Can Tell Us? From Correlation to Causation and Cognitive Ontologies, OHBM 2016 June 27, 2016 H

slide-1
SLIDE 1

Symposium “What Neuroimaging Can Tell Us? From Correlation to Causation and Cognitive Ontologies”, OHBM 2016 June 27, 2016

H       ?

Sebastian Weichwald MPI for Intelligent Systems

sweichwald.de/ohbm2016

  • www.neural.engineering
slide-2
SLIDE 2

Motivation

slide-3
SLIDE 3

Voxel weight for person.n.01

  • 4

+4 Voxel not used

Decoding model Encoding model LH

anterior sup. anterior sup.

RH

EBA EBA OFA OFA FFA FFA

2

Thanks to Alexander Huth, Gallant lab, UC Berkeley

slide-4
SLIDE 4

Voxel weight for person.n.01

  • 4

+4 Voxel not used

Decoding model Encoding model LH

anterior sup. anterior sup.

RH

EBA EBA OFA OFA FFA FFA

√ √ √ √ relevance

2

Thanks to Alexander Huth, Gallant lab, UC Berkeley

slide-5
SLIDE 5

Voxel weight for person.n.01

  • 4

+4 Voxel not used

Decoding model Encoding model LH

anterior sup. anterior sup.

RH

EBA EBA OFA OFA FFA FFA

√ √ √ √ relevance

2

Thanks to Alexander Huth, Gallant lab, UC Berkeley

slide-6
SLIDE 6

Encoding models

3

slide-7
SLIDE 7

Encoding models

GLM difference in means ⋯

3

slide-8
SLIDE 8

Encoding models

GLM difference in means ⋯ “Associated with experimental condition?”

3

slide-9
SLIDE 9

Reichenbach’s Common Cause Principle

4

Reichenbach, 1956

slide-10
SLIDE 10

Reichenbach’s Common Cause Principle

associated stimulus voxel

4

Reichenbach, 1956

slide-11
SLIDE 11

Reichenbach’s Common Cause Principle

stimulus voxel

4

Reichenbach, 1956

slide-12
SLIDE 12

Reichenbach’s Common Cause Principle

stimulus voxel h

4

Reichenbach, 1956

slide-13
SLIDE 13

Reichenbach’s Common Cause Principle

stimulus voxel h stimulus voxel

4

Reichenbach, 1956

slide-14
SLIDE 14

Reichenbach’s Common Cause Principle

stimulus voxel h stimulus voxel stimulus voxel

4

Reichenbach, 1956

slide-15
SLIDE 15

Reichenbach’s Common Cause Principle

stimulus voxel h stimulus voxel stimulus voxel

4

Reichenbach, 1956

slide-16
SLIDE 16

Causal interpretation chart

5

Weichwald et al., NeuroImage, 2015

slide-17
SLIDE 17

Causal interpretation chart

Voxel relevant? Encoding Causal interpretation Stimulus-based √ ×

5

Weichwald et al., NeuroImage, 2015

slide-18
SLIDE 18

Causal interpretation chart

Voxel relevant? Encoding Causal interpretation Stimulus-based √ effect of S ×

5

Weichwald et al., NeuroImage, 2015

slide-19
SLIDE 19

Causal interpretation chart

Voxel relevant? Encoding Causal interpretation Stimulus-based √ effect of S × no effect of S

5

Weichwald et al., NeuroImage, 2015

slide-20
SLIDE 20

Decoding models

6

slide-21
SLIDE 21

Decoding models

6

slide-22
SLIDE 22

Decoding models

“Does removal impair decoding performance?”

6

slide-23
SLIDE 23

Decoding models

“Does removal impair decoding performance?” “Associated after taking all other variables into account?”

6

slide-24
SLIDE 24

Causal interpretation chart

Voxel relevant? Encoding Causal interpretation Stimulus-based √ effect of S × no effect of S

7

Weichwald et al., NeuroImage, 2015

slide-25
SLIDE 25

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ ×

7

Weichwald et al., NeuroImage, 2015

slide-26
SLIDE 26

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ ×

Non-effects may appear relevant: stimulus voxel A voxel B

7

Weichwald et al., NeuroImage, 2015

slide-27
SLIDE 27

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ ×

7

Weichwald et al., NeuroImage, 2015

slide-28
SLIDE 28

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive ×

7

Weichwald et al., NeuroImage, 2015

slide-29
SLIDE 29

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive ×

Effects may be missed: stimulus voxel A voxel B

7

Weichwald et al., NeuroImage, 2015

slide-30
SLIDE 30

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive ×

7

Weichwald et al., NeuroImage, 2015

slide-31
SLIDE 31

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive × inconclusive

7

Weichwald et al., NeuroImage, 2015

slide-32
SLIDE 32

Are decoding models useful?

slide-33
SLIDE 33

Combine encoding and decoding models

&

8

slide-34
SLIDE 34

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive × inconclusive

9

Weichwald et al., NeuroImage, 2015

slide-35
SLIDE 35

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive × inconclusive √ √ √ × × √ × ×

9

Weichwald et al., NeuroImage, 2015

slide-36
SLIDE 36

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive × inconclusive √ √ √ × × √ × ×

Encoding plus decoding allows to identify indirect effects: stimulus voxel A voxel B

9

Weichwald et al., NeuroImage, 2015

slide-37
SLIDE 37

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive × inconclusive √ √ √ × × √ × ×

9

Weichwald et al., NeuroImage, 2015

slide-38
SLIDE 38

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive × inconclusive √ √ √ × indirect effect of S × √ × ×

9

Weichwald et al., NeuroImage, 2015

slide-39
SLIDE 39

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive × inconclusive √ √ √ × indirect effect of S × √ × ×

Encoding clarifies role of voxels relevant in decoding: stimulus voxel A voxel B

9

Weichwald et al., NeuroImage, 2015

slide-40
SLIDE 40

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive × inconclusive √ √ √ × indirect effect of S × √ × ×

9

Weichwald et al., NeuroImage, 2015

slide-41
SLIDE 41

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive × inconclusive √ √ effect of S √ × indirect effect of S × √ provides context × ×

9

Weichwald et al., NeuroImage, 2015

slide-42
SLIDE 42

Causal interpretation chart

Voxel relevant? Encoding Decoding Causal interpretation Stimulus-based √ effect of S × no effect of S √ inconclusive × inconclusive √ √ effect of S √ × indirect effect of S × √ provides context × × no effect of S

9

Weichwald et al., NeuroImage, 2015

slide-43
SLIDE 43

Wrap-up

slide-44
SLIDE 44

Wrap-up

▸ Simple causal interpretation chart

10

slide-45
SLIDE 45

Wrap-up

▸ Simple causal interpretation chart (also response-based paradigms)

10

slide-46
SLIDE 46

Wrap-up

▸ Simple causal interpretation chart (also response-based paradigms)

Relevant in encoding/correlated? ×/√ (e. g. GLM)

10

slide-47
SLIDE 47

Wrap-up

▸ Simple causal interpretation chart (also response-based paradigms)

Relevant in encoding/correlated? ×/√ (e. g. GLM) Relevant in decoding/partially correlated? ×/√ (e. g. classifier)

10

slide-48
SLIDE 48

Wrap-up

▸ Simple causal interpretation chart (also response-based paradigms)

Relevant in encoding/correlated? ×/√ (e. g. GLM) Relevant in decoding/partially correlated? ×/√ (e. g. classifier) ↝ Read off causal interpretation

10

slide-49
SLIDE 49

Wrap-up

▸ Simple causal interpretation chart (also response-based paradigms)

Relevant in encoding/correlated? ×/√ (e. g. GLM) Relevant in decoding/partially correlated? ×/√ (e. g. classifier) ↝ Read off causal interpretation

(Weichwald et al., NeuroImage, 2015)

10

slide-50
SLIDE 50

Wrap-up

▸ Simple causal interpretation chart (also response-based paradigms)

Relevant in encoding/correlated? ×/√ (e. g. GLM) Relevant in decoding/partially correlated? ×/√ (e. g. classifier) ↝ Read off causal interpretation

(Weichwald et al., NeuroImage, 2015)

Extensions

10

slide-51
SLIDE 51

Wrap-up

▸ Simple causal interpretation chart (also response-based paradigms)

Relevant in encoding/correlated? ×/√ (e. g. GLM) Relevant in decoding/partially correlated? ×/√ (e. g. classifier) ↝ Read off causal interpretation

(Weichwald et al., NeuroImage, 2015)

Extensions

▸ Stimulus-based causal inference

10

slide-52
SLIDE 52

Wrap-up

▸ Simple causal interpretation chart (also response-based paradigms)

Relevant in encoding/correlated? ×/√ (e. g. GLM) Relevant in decoding/partially correlated? ×/√ (e. g. classifier) ↝ Read off causal interpretation

(Weichwald et al., NeuroImage, 2015)

Extensions

▸ Stimulus-based causal inference

↝ Cause-effect relationships between neural processes

10

slide-53
SLIDE 53

Wrap-up

▸ Simple causal interpretation chart (also response-based paradigms)

Relevant in encoding/correlated? ×/√ (e. g. GLM) Relevant in decoding/partially correlated? ×/√ (e. g. classifier) ↝ Read off causal interpretation

(Weichwald et al., NeuroImage, 2015)

Extensions

▸ Stimulus-based causal inference

↝ Cause-effect relationships between neural processes

(Grosse-Wentrup et al., NeuroImage, 2016)

10

slide-54
SLIDE 54

Caveats

11

slide-55
SLIDE 55

Caveats

▸ Technical assumption of faithfulness

11

slide-56
SLIDE 56

Caveats

▸ Technical assumption of faithfulness

↝ “Nature does not hide dependencies.”

11

slide-57
SLIDE 57

Caveats

▸ Technical assumption of faithfulness

↝ “Nature does not hide dependencies.”

▸ Readiness to interpret negative results, i. e., uncorrelatedness

11

slide-58
SLIDE 58

▸ Causal and anti-causal learning in pattern recognition for

  • neuroimaging. PRNI, 2014. e-print arxiv.org/pdf/1512.04808.

▸ Causal interpretation rules for encoding and decoding models in

  • neuroimaging. NeuroImage, 2015. sweichwald.de/neuroimage2015.

▸ Identification of causal relations in neuroimaging data with latent

confounders: An instrumental variable approach. NeuroImage,

  • 2016. e-print mlin.kyb.tuebingen.mpg.de/Grosse-WentrupNI2015.pdf.

▸ Recovery of non-linear cause-effect relationships from linearly mixed

neuroimaging data. PRNI, 2016. e-print arxiv.org/pdf/1512.04808.

▸ MERLiN: Mixture Effect Recovery in Linear Networks. Under

  • review. e-print arxiv.org/pdf/1512.01255.

12

sweichwald.de/ohbm2016

N C

www.neural.engineering