Neural network study reveals small-world topology boosts polychronous group formation by 90%

A new study from Carneiro, Jiofack, and Ferreira published on arXiv (June 2026) dives into the structural determinants of Polychronous Neuronal Groups (PNGs) – reproducible, time-locked spatiotemporal firing cascades that rely on Spike-Timing-Dependent Plasticity (STDP) and heterogeneous axonal delays. The team simulated a recurrent network of 1,000 Izhikevich neurons over ten hours of biological time, detecting 1,545 unique PNGs using an offline event-driven algorithm. By sweeping the Watts-Strogatz topology from ring-lattice to random graph, they found that the clustering coefficient C is the primary driver: a ring-lattice (C~0.35) produced about 850 PNGs, while a random graph (C~0.20) generated fewer than 50 – a >90% drop in representational capacity.

Beyond topology, the researchers introduced a sparse-dot-product Recurrence Plot (RP) framework that identifies PNGs as unit-slope diagonal structures in the phase-space recurrence matrix, completely independent of anatomical neuron labeling. Recurrence Quantification Analysis yielded DET~0.65, quantifying reproducibility of the network’s dynamical trajectory. Together, these results establish small-world topology as the structural optimum for polychronization, and position the RP decoder as a principled, label-free tool for PNG identification. This work has implications for neuromorphic computing and understanding how neural networks encode temporal patterns.