Swarm AI optimizes animal brain networks for 17x better memory

Reservoir computing uses fixed recurrent neural networks for temporal processing, with only a linear readout trained. Biological connectomes—shaped by evolution—may encode computational structure beyond random networks, but whether that structure can be further optimized was an open question. A team led by Anmol Guragain tested this by applying four gradient-free, swarm intelligence optimizers (Particle Swarm, Differential Evolution, Grey Wolf, and Whale Optimization Algorithms) to the edge weights of connectome-based echo-state networks across six species spanning six orders of magnitude in neural complexity: C. elegans (279 neurons), Drosophila (49 nodes), mouse (112), rat (73), macaque (29 regions), and human (83 cortical parcels).

Each connectome was evaluated on four canonical reservoir computing benchmarks: Memory Capacity (MC), Lorenz attractor prediction, NARMA-10, and Mackey-Glass chaotic time-series. All four optimizers consistently outperformed unoptimized biological baselines across every task and species. The Whale Optimization Algorithm (WOA) achieved the largest gains: up to a 17x increase in memory capacity (C. elegans: from 1.39 to 23.91) and up to 89% reduction in normalized root mean square error (NRMSE) on the Mackey-Glass task with the human connectome—an average improvement of 214% across all species and tasks.

Crucially, when the optimizers were initialized with random weights (on the same topology), they reliably underperformed biology. This establishes biological weight values as an essential inductive bias that topology alone cannot recover. The results position biologically-initialised, swarm-optimised reservoir computing as a principled and effective strategy across the animal kingdom, suggesting evolution has left significant computational headroom for further optimization.