Brain-like features of MemComputing machines

TL;DR

This paper investigates brain-like features in MemComputing machines, emphasizing time non-local memory and phase-space dynamics as core mechanisms. It introduces the Universal MemComputing Machine (UMM) and its digital variant (DMMs), detailing intrinsic parallelism, asynchronous computation, information overhead, functional polymorphism, and analog/digital operability. It demonstrates how DMMs implement short- and long-term memories, Navier–Stokes-like instanton trajectories, and scale-free avalanche dynamics akin to cortical activity, providing robustness to noise while highlighting sensitivity to topological changes. The work suggests that such brain-inspired, emergent properties could inform hardware design and offer new perspectives for computational neuroscience, including insights into brain dynamics and developmental differences.

Abstract

MemComputing is a new model of computation that exploits the non-equilibrium property-we call 'memory'-of any physical system to respond to external perturbations by keeping track of how it has reacted at previous times. Its digital, scalable version maps a finite string of symbols into a finite string of symbols. In this paper, I will discuss some analogies of the operation of MemComputing machines-in general, and digital in particular-with a few physical properties of the biological brain. These analogies could be a source of inspiration to improve on the design of these machines. In turn, they could suggest new directions of study in (computational) neuroscience.