Dynamics of neural networks

Abstract

Animals are constantly submitted to a bombardment of information through their sensory systems. This information is transmitted to the central nervous system (CNS) in the form of spike trains. Traditional views of how this information is processed by the CNS consist in a series of networks (or layers) of neurons, connected in a predominantly feedforward manner. However, neurons in any cortical network receive their inputs not only from the previous layers (i.e., LGN for V1 neurons, V1 for V2 neurons, etc.), but also from nearby neurons that are part of the same network (“lateral” or “recurrent " connections), and from neurons in higher areas in the feedforward hierarchy (“top-down” connections). In fact, anatomy shows that feedforward inputs are typically a small minority of the inputs received by a cortical neuron (Binzegger et al. 2004). Therefore, to understand how networks of neurons in the CNS transmit the information that they receive, it is not enough to understand the input/output transformation at the single neuron level. In addition, one has to understand how the dynamics of networks of neurons shape the response of the population as a whole to dynamic inputs.