Hierarchical Shallow Predictive Matter Networks (HSPMN)

Summary

HSPMN is a research architecture that combines hierarchical predictive modules and parallel shallow processing with non‑reciprocal connectivity and oscillatory coordination. This repository provides the core specification and references.

Runnable comparison examples are available here: NetBr3ak/HSPMN-Examples.

References and context

• Predictive coding (Friston; Rao & Ballard)
• Shallow brain hypothesis (Suzuki et al.)
• Active matter dynamics (Marchetti; Cichos et al.)

Architecture (high level)

• Hierarchical predictive modules
• Parallel shallow processing units
• Dynamic non‑reciprocal connectivity
• Oscillatory coordination

Scope and status

• Goal: minimal, inspectable definition suitable for reproducible comparisons.
• Claims are limited to observed behavior in controlled tasks (see the examples repository). No performance guarantees beyond the tested settings.

Artifacts

• Paper: PDF/LaTeX (see repository files)
• Code: PyTorch reference implementation
• License: MIT (code); CC‑BY where applicable (text/figures)

Authorship

Szymon Jędryczko; with tooling assistance.