Controlling Integration and Segregation in Echo State Networks via Noradrenaline and Acetylcholine Neuromodulation

Abstract Biological brain networks flexibly reconfigure functional connectivity between integration and segregation through neuromodulatory systems—noradrenaline (NA) and acetylcholine (ACh)—without altering structural connectivity. Inspired by this mechanism, we propose a modular echo state network (ESN) with context-dependent NA and ACh gain modulation, where NA promotes inter-module integration via response gain and ACh promotes intra-module segregation via multiplicative gain. We evaluate the model on two context-dependent tasks: a segregation/integration task and a context-dependent decision task. Across both tasks, the modulated model consistently outperformed the baseline, with task-appropriate modulation profiles emerging naturally from optimization and functional connectivity analysis confirming context-appropriate dynamic reorganization. These results demonstrate that neuromodulatory gain control enables adaptive, context-sensitive computation in structurally fixed reservoir networks.