Research map / Architectures and efficiency / Spiking networks

From Cortical Synchronous Rhythm to Brain Inspired Learning Mechanism: An Oscillatory Spiking Neural Network with Time-Delayed Coordination

NeurIPS 2026 · Architectures and efficiency › Spiking networks · approach: Bio-inspired spiking neuron models · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes S²-Net, a spiking neural network coupled with a time-delayed Sakaguchi-Kuramoto synchronization gating mechanism, evaluated on brain fMRI state/disease classification and standard temporal SNN benchmarks.

Problem

Existing SNNs and Kuramoto-based models assume instantaneous coupling that favors zero-lag global synchrony, limiting flexible binding and temporal adaptation for brain-inspired computation.

Approach

Combines a rhythm-gated spiking neuron population with a vector-valued Sakaguchi-Kuramoto gating network using time-delayed phases and phase lags (estimated via Sinkhorn optimal transport on spike distributions) so that bottom-up spiking and top-down oscillatory synchronization interact iteratively.

Key results

• On HCP-WM (9-state temporal decoding), S²-Net reaches 88.72% accuracy versus 88.57% for Rhythm-SNN, a marginal gain with overlapping variance.
• On PS-MNIST, S²-Net achieves 96.94% accuracy vs 96.73% for Rhythm-SNN; on ECG it reports 89.23% vs 86.41%.
• On NIFD, S²-Net reports 64.65% accuracy vs 50.90% for Rhythm-SNN, the largest reported gap in the tables.

Surprising findings

• On HCP-WM and ADNI, S²-Net's accuracy is identical to or below Rhythm-SNN/DH-SNN in some cases, and the PPMI F1 (41.51) is below DH-SNN-level precision claims, suggesting uneven gains.
• Standard Transformer, LSTM and TCN baselines are competitive with or exceed SNN baselines on several neuroimaging datasets in the appendix.

Limitations

• Gains over strongest SNN baselines are small on several tasks and some differences are within reported standard deviations.
• Small neuroimaging cohorts (e.g., ADNI 135 subjects, PPMI 173) with high variance limit statistical confidence.
• Lyapunov stability analysis assumes slowly varying drive and symmetric coupling, while the model uses learnable phase lags and delays.

Experimental setup

Datasets: HCP-A HCP-YA HCP-WM UK Biobank ADNI PPMI NIFD S-MNIST PS-MNIST SHD ECG GSC

Benchmarks: HCP-A HCP-YA HCP-WM UK Biobank ADNI PPMI NIFD S-MNIST PS-MNIST SHD ECG GSC

Metrics: accuracy precision F1 MAE MSE

Baselines: FFSNN SRNN LSNN ASRNN DH-SNN Rhythm-SNN Transformer LSTM TCN Mamba AKOrN GraphCON

Builds on: Rhythm-SNN DH-SNN Kuramoto model Sakaguchi-Kuramoto model AKOrN

Closest papers by meaning

• A Scalable, Causal, and Energy Efficient Framework for Neural Decoding with Spiking Neural Networks (NeurIPS 2025 · similarity 0.88)
• Exploring the Limitations of Layer Synchronization in Spiking Neural Networks (NeurIPS 2026 · similarity 0.88)
• Learning the Plasticity: Plasticity-Driven Learning Framework in Spiking Neural Networks (NeurIPS 2025 · similarity 0.87)
• FiTS: Interpretable Spiking Neurons via Frequency Selectivity and Temporal Shaping (NeurIPS 2026 · similarity 0.87)
• BioOSS: A Bio-Inspired Oscillatory State System with Spatio-Temporal Dynamics (NeurIPS 2025 · similarity 0.87)
• SpikingSSMs: Learning Long Sequences with Sparse and Parallel Spiking State Space Models (AAAI 2025 · similarity 0.87)