Quantifying the energy dissipation and temporal irreversibility of the physical processes that occur in the brain is fundamental to understand its functioning. Despite the development of model-free estimators within the stochastic thermodynamics framework, minimal models remain essential for characterizing complex systems out of equilibrium.
In this talk, I discuss entropy production in the stochastic Wilson-Cowan model, a coarse-grained model describing the dynamics of large populations of excitatory and inhibitory neurons. In the linear noise approximation, the fluctuations of active excitatory and inhibitory populations are described by two coupled linear Langevin equations. Thus, the entropy production can be computed analytically allowing to better understand the role of the excitation-inhibition imbalance in sustaining non-equilibrium steady states. Beyond theoretical considerations, I also discuss the practical problem of characterizing the equilibrium state of the system from partial observations. For Gaussian processes, I prove that it is impossible to distinguish the thermodynamic state of the system by measuring a scalar variable without performing response experiments. In cases where it is not possible to carry out response experiments, I explain a possible strategy for inferring the non-equilibrium properties of the system by combining the information of different data sets acquired in different experimental condition.
Prof. Luca Salasnich
