Sparse coding of different tuning functions in the primary visual cortex Position Eye (stereo image) Spatial frequency (scale) Orientation Direction and speed of motion Wavelength (color) Courtesy of Aapo Hyvärinen
Visual information Correlated features Sparse coding Independent representations
Context supports perception
Context distorts perception
Area tuning function Varying size of drifting gratings Courtesy of Lauri Nurminen and Markku Kilpeläinen
Receptive field Angelucci & Bressloff, Prog Brain Res 154 (2006) 93 – 120
A block model of surround interaction Schwabe et al. J Neurosci 26 (2006) 9117-9129 Afferent input Low-level area High-level area
Subtractive normalization model applied to non-linear interactions in the human cortex What visual information has to do with surround modulation?
Stimuli Vanni & Rosenström, in preparation
Centre response covaries with the surround response Vanni & Rosenström, in preparation VOIcentre
Active voxels for centre are suppressed during simultaneous presentation Vanni & Rosenström, in preparation VOIcentre
Suppression (red) is surrounded by facilitation (blue) Vanni & Rosenström, in preparation
Efficient coding Response to stimulus A, A’ Response to stimulus B, B’ A’ = A – dB B’ = B – dA Barlow, H., and Földiák, P. (1989). In: The computing neuron. R. Durbin, et al., eds. (Boston, Addison-Wesley Longman Publishing Co., Inc), pp. 54-72.
Effective use of narrow dynamic range (surround modulation) and limited time (adaptation)
More explicit causal factors
Implemented by Hebbian and anti-Hebbian learning rules
Barlow, H., and Földiák, P. (1989). In: The computing neuron. R. Durbin, et al., eds. (Boston, Addison-Wesley Longman Publishing Co., Inc), pp. 54-72.
A hypothesis of the visual brain
Our brain learns a hierarchical model of our visual environment
Each neuron in the model is sensitive to a set of correlated features in the environment
Population of neurons in this model form a sparse representation by relatively independent units
The tuning functions may be the most informative dimensions of visual environment