DA-LIF: Dual Adaptive Leaky Integrate-and-Fire Model for Deep Spiking Neural Networks

TL;DR

DA-LIF tackles limited expressiveness in traditional LIF neurons by introducing independently learnable spatial and temporal membrane decays, realized as $\alpha^n$ and $\beta^n$ in the membrane-update $V^{t,n} = \beta^n H^{t-1,n} + \alpha^n X^{t,n}$. The method advances from a variable-decay LIF to a dual adaptive mechanism with separate time constants $\tau_a$ and $\tau_b$, enabling per-layer adaptation of both spiking and input integration. Trained with spatio-temporal backpropagation and surrogate gradients on static (CIFAR-10/100, ImageNet) and neuromorphic (CIFAR10-DVS, DVS128 Gesture) datasets, DA-LIF achieves state-of-the-art accuracy with fewer timesteps and minimal parameter overhead, while maintaining energy efficiency. Ablation studies show the complementary roles of spatial and temporal tuning, the advantage of the tanh activation, and a clear layer-wise specialization of the decays, underscoring robustness and practical impact for energy-efficient neuromorphic deployments.

Abstract

Spiking Neural Networks (SNNs) are valued for their ability to process spatio-temporal information efficiently, offering biological plausibility, low energy consumption, and compatibility with neuromorphic hardware. However, the commonly used Leaky Integrate-and-Fire (LIF) model overlooks neuron heterogeneity and independently processes spatial and temporal information, limiting the expressive power of SNNs. In this paper, we propose the Dual Adaptive Leaky Integrate-and-Fire (DA-LIF) model, which introduces spatial and temporal tuning with independently learnable decays. Evaluations on both static (CIFAR10/100, ImageNet) and neuromorphic datasets (CIFAR10-DVS, DVS128 Gesture) demonstrate superior accuracy with fewer timesteps compared to state-of-the-art methods. Importantly, DA-LIF achieves these improvements with minimal additional parameters, maintaining low energy consumption. Extensive ablation studies further highlight the robustness and effectiveness of the DA-LIF model.

Figures (3)

• Figure 1: Overview of the DA-LIF model and DA-LIF-SNN Architecture. (a) Experimental Motivation of the proposed method. (b) convention adaptive LIF model (c) proposed DA-LIF model (d) $\alpha^n$ and $\beta^n$ are independently learned and share parameters within the same layer, but differ across each layer in networks.
• Figure 2: Impact of $\alpha$ and $\beta$ on CIFAR-100 with ResNet-20
• Figure 3: Distribution of $\alpha$ and $\beta$ Across Layers.