Energetics, dynamics and structure of spiking neural networks under metabolic constraints

Abstract

Every biological tissue or physical system is subject to physical restrictions that limit its functioning. Specifically in neural networks, energy constraints determine physically feasible states in which they can evolve. However, this fundamental concept of energy has been largely overlooked when modeling and simulating the dynamics of neural networks. This thesis aims to formalize, study and simulate the dynamics and structure that emerges in spiking neural networks when there are local metabolic restrictions that affect behavior at the neuronal and synaptic level. In particular, through the creation of an energy dependent single-neuron model and an energy dependent plasticity rule, the impact generated by different types and intensities of energy constraints on connectivity and activity in an excitatory-inhibitory balanced network is studied both analytically and through simulation. When neurons and synapses are sensitive to energy imbalances, metabolic stable fixed points appear at the network level, which are mathematically described and validated through simulations. The developed framework allows the study of neural networks under impaired metabolic conditions. Therefore, the proposed theoretical and simulation framework introduced in this work could be valuable to deepen the knowledge about the relationship between neurodegenerative diseases and metabolic impairments at the neuronal, synaptic, and network levels.