Item

Abstract

Our sensory systems perform pretty successful and fast in processing the complex scenarios of sensory stimuli that we encounter during every day tasks. It seems likely that over evolutionary
timescales our senses adapted to the statistical regularities inherent to the every day stimuli
that we have to analyze. This leads to the hypothesis that various aspects of sensory processing
are ‘optimal’ for extracting information from natural input; vice versa, one should be able to
understand the organization of sensory processing given the statistical regularities of natural
stimuli and given the right optimization principle. In this talk I will address this question by
examining the processing of single neurons in the primary visual cortex, so called simple and
complex cells. Recent work showed how the processing of these cells can be understood as
efficient and redundancy reducing codes of natural visual features—called sparse and slow feature
codes. Especially, by applying these optimality principles to artificial networks processing
natural visual scenes, properties of simple and complex cells can be reproduced. Besides vision,
other senses perform similar analysis of complex stimuli, posing the question whether
different senses can be understood using similar functional principles. Using audition as an
example, it can be shown that the same optimality principles deduced from the visual cortex
do a good job in explaining the processing of auditory neurons. This suggests that, on a more
abstract level, different sensory systems perform similar operations that are defined by certain
optimality criteria and the statistical properties of natural sensory stimuli. Especially, the early
sensory processing in different sensory systems seems to be based on a small set of common
computational principles that links properties of the outside world to properties of neuronal
circuits.