Research map / Architectures and efficiency / Spiking networks

Neuronal Self-Adaptation Enhances Capacity and Robustness of Representation in Spiking Neural Networks

NeurIPS 2026 · Architectures and efficiency › Spiking networks · approach: Spiking transformers · 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 KvLIF, a LIF variant with an auxiliary potassium-conductance state that modulates subthreshold input shunting and post-spike reset, reporting accuracy and noise-robustness gains on static and neuromorphic SNN benchmarks.

Problem

Conventional LIF neurons have a narrow dynamic response range for varying input intensities and soft reset causes noise-induced false spikes, limiting SNN accuracy and robustness.

Approach

An auxiliary potassium conductance state K[t] integrates membrane potential and spiking history; it shunts subthreshold input (A-type-like) and deepens the post-spike reset (delayed-rectifier-like, after-hyperpolarization), extending dynamic range and suppressing noise spikes.

Key results

• KvLIF with Spiking ResNet-18 (T=6) reaches 95.55% on CIFAR10 and 79.30% on CIFAR100, versus 94.70% and 77.17% for LIF-H.
• On DVS-Gesture, KvLIF (VGG-11, T=10) achieves 98.26% accuracy, above the best listed baseline at 97.92%.
• Under CIFAR100 Gaussian noise at std 0.20, KvLIF gets 33.32% versus 28.05% for GLIF and 21.33% for LIF-H.

Surprising findings

• KvLIF attains over 90% accuracy on static data in just one time step according to the paper's stability analysis.
• The ablation indicates adaptive repolarization contributes more to performance than synaptic shunting.

Limitations

• Code is promised only after publication; no release at submission time.
• Hyperparameters (alpha, beta, gamma) are set per dataset, and comparisons use reported numbers from differing architectures and timesteps.

Experimental setup

Datasets: CIFAR10 CIFAR100 Tiny-ImageNet CIFAR10-DVS DVS-Gesture

Benchmarks: CIFAR10 CIFAR100 Tiny-ImageNet CIFAR10-DVS DVS-Gesture

Metrics: classification accuracy accuracy under Gaussian noise accuracy under temporal event dropping firing rate SOP energy

Baselines: LIF-S LIF-H GLIF CLIF ILIF Dspike DSR SML PLIF AR-LIF PSN AGMM

Builds on: LIF TET CLIF GLIF PLIF

Closest papers by meaning

• TC-LIF: A Two-Compartment Spiking Neuron Model for Long-Term Sequential Modelling (AAAI 2024 · similarity 0.88)
• CLIF: Complementary Leaky Integrate-and-Fire Neuron for Spiking Neural Networks (ICML 2024 · similarity 0.87)
• The Expressive Leaky Memory Neuron: an Efficient and Expressive Phenomenological Neuron Model Can Solve Long-Horizon Tasks. (ICLR 2024 · similarity 0.86)
• Rethinking the Membrane Dynamics and Optimization Objectives of Spiking Neural Networks (NeurIPS 2024 · similarity 0.86)
• Enhanced Self-Distillation Framework for Efficient Spiking Neural Network Training (NeurIPS 2025 · similarity 0.86)
• FEEL-SNN: Robust Spiking Neural Networks with Frequency Encoding and Evolutionary Leak Factor (NeurIPS 2024 · similarity 0.86)