Recent advances at the intersection of control theory, nonlinear dynamics, and machine learning have revealed novel mechanisms by which networked dynamical systems perform computation. In this talk, I extend this line of research to show that random heterogeneity in nodal parameters—often treated as disorder—can be exploited as a mechanism for stability and decision-making in neurocomputing systems. I first derive conditions under which the stability of desirable network-level states is promoted by disorder, revealing a direct link to the dimensionality of the underlying nodal dynamics. I then turn to decision-making systems, where I demonstrate that introducing disorder can significantly bias the system convergence toward (near-)global minimizers. These findings are supported through a combination of analytical results, numerical simulations, and experimental validation. The talk concludes by outlining future directions on energy-based dynamical models, with applications in associative memory, sparse optimization, and unconventional computing.
Speaker
Arthur MontanariPostdoctoral scholar at the Center for Network Dynamics at Northwestern University