Orban-Sloan-Swartz-summer-2007

Spontaneous activity in V1:

a probabilistic framework



Gergő Orbán

Volen Center for Complex Systems

Brandeis University









Sloan Swartz Centers Annual Meeting, 2007

Normative account for visual

representations

 Optimization criterion for the emergence of simple-cell

receptive fields: independent ‘filters’ + sparseness

(Bell & Sejnowski, 1995; Olshausen & Field, 1996)

Activity in V1

The spectrum of V1 physiology is much richer

 Spontaneous activity

 Response variabilty

 Temporal dynamics





Can we devise a framework that

 Gives a functional description of visual processing

 Uses normative principles in probabilistic learning

 Gives a more complete interpretation of V1

activity?

Computational paradigm

Density estimation

 Statistically well founded

principle

 Allows the representation

of uncertainty

 Efficient for making

predictions



Internal representation:

 Useful representation

 Biologically plausible

: retinal image/ RGC output; : neural activity

Spontaneous activity

 In the awake brain there is

patterned neural activity not

directly related to the stimulus



Evoked Spontaneous

 Patterns of neural activities

are similar in stimulus evoked Qu i ckTi m e™ a nd a

TIFF (LZW) de co mp res so r

a re ne ed ed to se e th is pi c tu re.

Qu i ckTi m e™ a nd a

TIFF (LZW) de co mp res so r

a re ne ed ed to se e th is pi c tu re.



condition and closed eye

condition

(Tsodyks et al, 1999)





 Long-range correlations in

neural activity





(Fiser et al, 2004)

Probabilistic model:

Field of experts

 Filters are componenets in a Receptive fields

Boltzmann energy function (Osindero,

Welling & Hinton, 2006)



 Sparse prior (Student-t distribution)



 Image model assuming translational

invariance (Black & Roth, 2005)



 Learning: standard contrastive

divergence & Hybrid MC (Hinton 2002)

Spontaneous activity as

prior sampling

Evoked activity:



ANSATZ:

Spontaneous

activity:



Evoked Natural

activity image

statistics

Intuitive link between evoked and

spontaneous activities

Images generated by the model



Prior over activities

Sampling



Neural activities

Filters

Dreamed image









Images generated from prior have long-range structure

Evoked and spontaneous neural

activity

Correlation between

hidden units

Experiment









(Fiser et al, 2004)



Evoked and spontaneous activities have similar

correlational structure

Spontaneous neural activity

before learning



Experiment









(Fiser et al, 2004)





Correlational patterns in the activity of neurons

is a result of learning in the probabilistic model

Conclusions

 The probabilistic framework provides a viable

explanation for spontaneous activity in V1

 Spontaneous activity as sampling from prior

 Long range correlations are present both in

evoked and spontaneous activities

 The tendency of changes in spatial correlations

with training match experimental results

Bottom line



In the probabilistic framework:



 Spontaneous activity

prior sampling

 Response variablity

posterior variance

 Temporal dynamics

top-down/

lateral interactions

Special thanks to

Pietro Berkes (Gatsby)



Collaborators:

Máté Lengyel (Gatsby)

József Fiser (Brandeis)

principles + physiology

High-level computational – prior sampling



• Computational paradigm: – posterior variance

– top-down/

Normative probabilistic model

lateral interactions

• Experimental paradigm:

Spontaneous activity in V1

Are there sensible

interpretations that assign

functional roles for the

spontaeous activity?