Scaling behavior of PHCSSM under larger hierarchical configurations

Determine the scaling behavior of the Parallelized Hierarchical Connectome State-Space Model (PHCSSM) when increasing the hierarchy and state size beyond the two-region, D = 16–64 setting evaluated in this paper, specifically for larger configurations such as 4 regions with 128 neurons per region (4R128) and 6 regions with 64 neurons per region (6R64), assessing effects on accuracy, stability, and computational efficiency.

Background

The paper evaluates PHCSSM on six UEA physiological benchmarks using modest state dimensions (D = 16–64) and a two-region hierarchy. While results are competitive and parameter-efficient, the authors note that the architecture has not been tested under larger hierarchical settings or higher state sizes.

Understanding how performance, stability, and efficiency scale with the number of regions and neurons per region is essential to assess PHCSSM’s applicability to larger tasks and connectome-scale settings.

References

The current evaluation is restricted to binary and multi-class classification at modest state dimensions (D = 16--64) with a two-region hierarchy; scaling behavior under larger configurations (4R128, 6R64) remains unknown.

— Parallelized Hierarchical Connectome: A Spatiotemporal Recurrent Framework for Spiking State-Space Models  (2604.01295 - Chiang, 1 Apr 2026) in Discussion — Limitations