a very short introduction to multivariate pattern analysis (mvpa) … · 2014. 11. 10. · a very...

A very short introduction to multivariate patternanalysis (MVPA) for neuroscience

Michael Hanke & Yaroslav Halchenko

University of Magdeburg, GermanyDartmouth College, USA

Giessen 2014

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 1 / 16

SPM via GLM (“Encoding model”)

p(brain activity|behavior)

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 2 / 16

SPM via GLM (“Encoding model”)

?ResearchQuestion

ExperimentDesign

Stimuli

GLMGLMGLM

HypothesisTesting

GLM

SPM

??

?

?

? ?

p(brain activity|behavior)

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 2 / 16

That’s not enough

Eric Kandel in Principles of Neuroscience

“The task of neural science is to explain behavior in terms ofthe activities of the brain.”

p(brain activity|behavior) 6= p(behavior|brain activity)

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 3 / 16

Approach: Meta analysis and Bayes’ theorem

Yarkoni et al., Nature Methods, 2011; http://neurosynth.orgH2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 4 / 16

Why multivariate methods?

Haufe et al., 2014, NeuroImageH2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 5 / 16

Why multivariate methods?

−4 −2 0 2 4

−4

−2

02

4

●● ●● ●●● ●● ●● ●●● ● ●●● ●● ●● ●● ● ●● ●● ●● ●● ●●● ●● ●● ● ●● ●● ● ●●● ●

−4 −2 0 2 4

● ●●● ●● ●●● ●●● ●●● ● ●●●● ● ● ● ●●●●● ● ●● ● ●● ●● ●●● ● ●● ● ●●● ●● ●●

−4 −2 0 2 4

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 6 / 16

Why multivariate methods?

−4 −2 0 2 4

−4

−2

02

4

●● ●● ●●● ●● ●● ●●● ● ●●● ●● ●● ●● ● ●● ●● ●● ●● ●●● ●● ●● ● ●● ●● ● ●●● ●

−4 −2 0 2 4

● ●●● ●● ●●● ●●● ●●● ● ●●●● ● ● ● ●●●●● ● ●● ● ●● ●● ●●● ● ●● ● ●●● ●● ●●

−4 −2 0 2 4

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 6 / 16

Pioneering work: visual objects

Haxby et al., Science, 2001H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 7 / 16

k-Nearest Neighbours

−4 −2 0 2 4

−4

−2

02

4

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 8 / 16

k-Nearest Neighbours

−4 −2 0 2 4

−4

−2

02

4

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 8 / 16

k-Nearest Neighbours

−4 −2 0 2 4

−4

−2

02

4

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 8 / 16

k-Nearest Neighbours

−4 −2 0 2 4

−4

−2

02

4

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 8 / 16

MVPA approach: Reverse the flow!

p(behavior|brain activity)H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 9 / 16

Which classifier?

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 10 / 16

Decision models – linear problem

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 11 / 16

Decision models – non-linear problem

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 11 / 16

Model appropriateness

James, G., Witten, D., Hastie, T., and Tibshirani, R. (2013). An Introduction to Statistical Learning:with Applications in R. Springer Texts in Statistics. Springer. (free PDF copy)

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 12 / 16

Data representation – classification

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 13 / 16

Enough said. . . (for now)

Let’s see how we can do this. . .

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 14 / 16

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 15 / 16

References

James, G., Witten, D., Hastie, T., and Tibshirani, R. (2013). An Introduction to Statistical Learning: with Applications in R.Springer Texts in Statistics. Springer. (free PDF copy).

H2 (Dartmouth; Magdeburg) MVPA Intro Giessen 2014 16 / 16