Our goal is to better understand how the brain processes visual stimuli such as natural images or natural image sequences.

We focus on the area V4 of the brain, which is known to respond to visual features of medium complexity such as simple shapes. The data is provided by Jack Gallant (UC Berkeley) and consists of recordings from electrodes in the V4 area of awake subjects.

Following the methodology of Kay et al. 2008 , we address prediction tasks:

• given a visual stimuli, can we predict the corresponding brain activity? (encoding)
• given an electrode recording, which information can we infer about the original visual stimuli? (decoding)

The first difficulty of this approach is to find a good image representation that

• is adapted to the statistics of natural images.
• has similar invariant properties as the V4 area to visual stimuli.