Predictive coding in the early visual system

I have investigated the strategies that neural systems employ to adapt to the diversity of their input and code it in an informationally efficient way.

In particular, I have studied how neurons in the early visual system code the broad statistics of natural visual input into their relatively narrow dynamic range of activity.

We show (Buiatti and van Vreeswijk, 2003) that neurons can maximize information transmission by reducing temporal correlations present in the visual input through variance normalisation. This predictive coding model is inspired by experimental results on the early visual system of the fly and is trained on time series recorded in a natural environment.

References:

Buiatti M, Van Vreeswijk C, 

Variance normalisation: a key mechanism for temporal adaptation in natural vision?, 

Vision Research 43 (17), 1895-1906 (2003).