Prof. Alessandro Torcini
(LPTM, Université de Cergy-Pontoise, France)
During this advanced study group we will address two interlinked questions: encoding/decoding and transport of information in neural networks. How are microscopic, mesoscopic and macroscopic scales bridged together to achieve a robust information processing, starting from the single neuron activity?
Information coding lies at the heart of a strong debate with various strategies being considered by different research groups. The principal options include temporal, rate and population coding. Without pretending to give a final answer to this question, we will focus our analysis on the limits and potentialities of population coding as it allows for an easier investigation of its role over different scales. It is known that neural circuits are subject to strong fluctuations due to internal noise and to inputs arriving from distant, possibly external, areas. Furthermore, recent experimental results pointed out that neural circuits may well be chaotic, i.e. dynamically unstable. How is it then possible to unambiguously retrieve the information content of a given stimulus from the dynamics of a neural circuit? This question is fundamental for the neuroscientists? community, but it is also of extreme interest for the researchers working on nonlinear dynamics and complex systems. One possible answer is population coding: it assumes that information is encoded in the (ensemble) averaged behavior of a given subpopulation, so that the irrelevant stochastic component is smoothed out.
As for the propagation of information in a neural network, the starting point is the need to go beyond the linearization of the equations of motion. The unavoidable presence of noise makes the evolution of perturbations on tiny observational scales rather uncontrollable. It is nowadays clear that new indicators of information propagation and transfer need to be considered in neural circuits which relate to finite scales both at the spatial and temporal level. Therefore the usual indicators employed to quantify information transport, such as the Kolmogorov-Sinai entropy, the generalized Renyi entropies, and the Lyapunov spectra, are probably only of limited use. This advanced study group will focus on the development of new indicators capable to characterize information processing at a finite resolution.
A system where the strategies associated to population coding and information transfer can be realistically tested is the striatal medium spiny network, a neural circuit made of sub-networks of inhibitory interneurons, which is crucial for both motor control and learning.