| name | hierarchical-connectome-phc |
| description | Parallelized Hierarchical Connectome (PHC) framework for spatiotemporal recurrent spiking neural networks. Upgrades State-Space Models (SSMs) into spatiotemporal networks with biological constraints including Dale's Law, short-term plasticity, and reward-modulated STDP. Activation: spiking neural networks, SSM, connectome, spatiotemporal modeling, biological neural networks. |
Parallelized Hierarchical Connectome (PHC)
Overview
This work presents the Parallelized Hierarchical Connectome (PHC), a general framework that upgrades temporal-only State-Space Models (SSMs) into spatiotemporal recurrent networks by mapping SSM components to hierarchical neuronal architectures with biological constraints.
Paper Reference
- Title: Parallelized Hierarchical Connectome: A Spatiotemporal Recurrent Framework for Spiking State-Space Models
- arXiv ID: 2604.01295v1
- Authors: Po-Han Chiang
- Published: 2026-04-01
- Category: q-bio.NC (Neurons and Cognition)
- PDF: https://arxiv.org/pdf/2604.01295v1
Core Innovation
PHC maps the diagonal SSM core to a shared Neuron Layer and inter-neuronal communication to a shared Synapse Layer, where neurons are partitioned into hierarchical regions governed by the connectome topology.
Key Features
-
Multi-Transmission Loop: Enables intra-slice spatial recurrence within each temporal window while preserving O(logT) parallelism
-
Biological Constraints: Supports intractable neuro-physical priors including:
- Adaptive leaky integrate-and-fire (ALIF) dynamics
- Dale's Law (excitatory/inhibitory neuron separation)
- Short-term plasticity
- Reward-modulated spike-timing-dependent plasticity (STDP)
-
Parameter Efficiency: Reduces complexity from Θ(D²L) for L-layer stacked architectures to Θ(D²)
PHCSSM Implementation
PHCSSM is the first model to unify:
- Recurrent spiking neural network dynamics
- Diagonal SSM parallelism
- Five biological constraints
- Learnable lateral connections
- Fully parallelizable training pipeline
Architecture Components
Neuron Layer (NL)
Maps diagonal SSM core to hierarchical neuronal regions
Synapse Layer (SL)
Inter-neuronal communication with connectome topology
Multi-Transmission Loop
- Intra-slice spatial recurrence
- Preserves parallel scan efficiency
- Enables lateral/feedback interactions within single timestep
Biological Constraints Supported
| Constraint | Description |
|---|
| ALIF | Adaptive leaky integrate-and-fire dynamics |
| Dale's Law | Excitatory/inhibitory neuron separation |
| STP | Short-term plasticity |
| R-STDP | Reward-modulated spike-timing-dependent plasticity |
Empirical Results
Evaluated on physiological benchmarks from the UEA multivariate time-series archive:
- Performance: Competitive with state-of-the-art SSMs
- Parameter complexity: Reduced from Θ(D²L) to Θ(D²)
- Training: Fully parallelizable pipeline
Methodology Applications
Use this framework when:
- Building biologically grounded sequence models
- Implementing spiking neural networks with SSM efficiency
- Researching brain-inspired parameter-efficient architectures
- Studying spatiotemporal dynamics in neural systems
Trigger Keywords
- spiking neural network
- state-space model
- connectome
- spatiotemporal recurrence
- biological neural network
- Dale's Law
- short-term plasticity
- reward-modulated STDP
- parallel scan
- parameter-efficient SSM
