Abstract
<title>Abstract</title> <p>Understanding how network structure emerges from functional demands is a central principle in network physiology and computational neuroscience. A key question remains whether compact computational models can recover the \textit{structure-function} coupling of neuronal systems from dynamics alone, without anatomical supervision. We apply the performance-dependent network evolution (PDNE) framework to model the Wilson-Cowan (WC) neuronal system, a canonical two-population model of excitatory-inhibitory (E-I) interaction underlying physiological rhythms. Starting from minimal seed networks of varying sizes, PDNE iteratively grows and prunes a reservoir computing network under explicit prediction-performance pressure, producing a compact model whose emergent topology reflects the target system’s structure. The evolved networks accurately predict both excitatory $E(t)$ and inhibitory $I(t)$ population activities across unseen stimulus amplitudes and generalize in a zero-shot manner to novel stimulus configurations without retraining. Mechanistically, PDNE achieves accurate dynamics through a fundamental transition from constrained low-dimensional activity to organized higher-dimensional attractor exploration, accompanied by strategic reorganization of nodes into specialized populations. Structural analysis reveals consistent functional organization across networks of final sizes, with the population-level connectivity spontaneously recovering the correct excitatory-inhibitory sign pattern of the WC model for three of four interaction types, without this being imposed by design. These results instantiate network physiology principles at the neuronal subsystem scale that performance-driven network evolution produces not only accurate but structurally interpretable models, opening a path toward compact, mechanistically grounded and data-efficient digital twins of neuronal systems.</p>