Stochastic modeling of biological neural networks and animal behavior

par Dr Guilherme Ost (Institute of Mathematics of the Federal University of Rio de Janeiro)

jeudi 16 mars 2023, 14:00 → 15:00 Europe/Paris

A central question in neuroscience is to understand how large-scale brain dynamics emerge from the interactions between individual neurons. A popular approach for addressing this question is to consider large-scale limits of interacting point processes modeling the neural activity, assuming that the interaction between the components of the model is of mean-field type (“uniform”). However, as I will show in the talk, in such mean-field limits the evolution of any pair of components (the neurons) become independent, contrarily to what is often observed in empirical data. This leads naturally to the following question: if neurons do not interact in a mean-field way, how do they interact? A crucial step to answer this question is to design methods to reconstruct the potential interactions between neurons from their spiking activity. Here I will briefly show some statistical methods to neural interaction reconstruction that I have proposed with some collaborators, discussing also some of their limitations. After that, in the remaining part of my talk,I will present a random walk model that modifies the dynamics depending on its trajectory, motivated by the following question: why did we observe just a few behavioral states in animals, although they typically have a large degree of freedom for producing movements?The standard approach for the predominance of a subset of behavior in an individual is to assume a learning mechanism. Here, I will show that it is also possible to observe only a few dominant behaviors, despite the lack of learning.