Title:Frenetic steering in a nonequilibrium graph

Abstract:In traditional recognition tasks of neural networks, a potential landscape or cost function guides the system toward patterns using gradient dynamics. That is not how the brain works as its dynamics is far from equilibrium. We present an alternative and proof of principle for pattern recovery in a nonequilibrium model whereby only time-symmetric kinetics are altered. As a mathematical model, a random walker on a randomly-oriented complete graph is subject to finite driving in the direction of the arcs. Some vertices of the graph represent patterns. A first algorithm constructs basins of attraction for these patterns. A second algorithm updates the time-symmetric factors in the transition rates, in order for the walker to quickly reach a pattern and remain there for a sufficiently long time, whenever starting from a vertex in its basin of attraction.