Information processing organization of a biological neural network

Abstract

Reservoir computing has been recently proposed as a paradigm of how the brain processes time-dependent complex information. It relies on a recurrent core of neurons, known as the reservoir, that receives and encodes complex inputs in a highdimensional phase space. Encoding takes place by combining the incoming input signal with the existing state of the network, which depends on past inputs. This provides this computational paradigm with its ability to process a temporally varying environment. While this concept was proposed a few years ago as a potential mechanism of information processing by the brain, and in spite of the overwhelming evidence of recurrent connectivity found in the brain, it has been difficult to validate this hypothesis given the extreme structural complexity of the mammalian brain. The aim of this thesis is to study how information is propagated within a biological neural network, using the connectome of C. elegans as a model system. This study may help to characterize possible biological network architectures such as Reservoir Computing that explain cognitive behaviors such as thermotaxis. This paradigm could help in the understanding of how a basic nervous system works, for further inferences in higher complex systems.