The $α$-Cap Process: A Continuous Model for Random Geometric Networks of Binary Neurons

Recurrent networks of binary neurons are a foundational concept in artificial intelligence. While these networks are traditionally assumed to be fully connected, complex dynamics can emerge when the graph structure is varied. One graph structure of particular interest is the geometric random graph, which models the spatial dependencies present in biological neural networks. In such classes of graphs, global state dependencies tend to complicate analysis, motivating the study of their dynamics in the continuum limit. In this work, we propose and analyze a continuous model for the evolution of binary neuron states in $\mathbb{R}^d$ via a function $ψ:\mathbb{R}^d\to[0,1]$ encoding the neural activity at a point. Our analysis encompasses a class of processes defined by convolution and sharpening; we demonstrate that, when evolved this process, the level sets of $ψ$ asymptotically converge to balls in $\mathbb{R}^d$. Notably, a special case of this process is the Merriman-Bence-Osher (MBO) scheme for the motion of interfaces by mean curvature[MBO92], and we provide a novel analysis of its behavior. Our results establish a surprising connection between geometric random graphs and classical models of interface motion, offering new insights into the interplay between spatial structure and neural dynamics.