Is Free Energy an Organizational Principle in Spiking Neural Networks?

JOSÉ A. FERNÁNDEZ LEÓN; MARCELO ARLEGO; ACOSTA, GERARDO G.

LECTURE NOTES IN COMPUTER SCIENCE

Springer Nature

Lugar: Zurich; Año: 2022 p. 79 - 90

0302-9743

An open question in neuroscience is how the brain self-organizes. Despite significant progress in such understanding, this effort is still difficult to address in biological neural networks due to limitations in recording all the involved network components. It is possible however to approach this issue by examining how relatively small bio-inspired networks self-organize its neural activities under different states. From a computational standpoint, one can have access in this way to all variables and parameters of the network. Here we dis- cuss how (non-variational) free energy varies during drastic changes in a spiking network through a sleep-like transition and in the absence of sensing. This non- variational free energy is defined as in its thermodynamic counterpart, where ‘energy’ refers to a quantity whose mean value is known or estimated from data sampling or model generated. We propose a novel view of how free energy is organized in a cortical-like neural network differently according to the synchro- nization of the network under external or internal fluctuations. Empirically testable hypotheses are presented in the context of non-sensory free energy modulation.