Spiking Neural Models of Neurons and Networks for Perception, Learning, Cognition, and Navigation: A Review

This article reviews and synthesizes highlights of the history of neural models of rate-based and spiking neural networks. It explains that theoretical and experimental results about how all rate-based neural network models, whose cells obey the membrane equations of neurophysiology, also called shunting laws, can be converted into spiking neural network models without any loss of explanatory power, and often with gains in explanatory power. These results are relevant to all the main brain processes, including individual neurons and networks for perception, learning, cognition, and navigation. The results build upon the hypothesis that the functional units of brain processes are spatial patterns of cell activities, or short-term-memory (STM) traces, and spatial patterns of learned adaptive weights, or long-term-memory (LTM) patterns. It is also shown how spatial patterns that are learned by spiking neurons during childhood can be preserved even as the child’s brain grows and deforms while it develops towards adulthood. Indeed, this property of spatiotemporal self-similarity may be one of the most powerful properties that individual spiking neurons contribute to the development of large-scale neural networks and architectures throughout life.