Advances in Spiking Neural Networks

The field of Spiking Neural Networks (SNNs) is rapidly advancing with a focus on improving their performance, efficiency, and applicability to various tasks. Recent developments have shown that SNNs can be effectively used for image deraining, object detection, and emotion recognition, among other tasks. The introduction of new encoding schemes, such as learnable temporal encoding and hybrid temporal-bit spike coding, has enhanced the ability of SNNs to model complex temporal dynamics. Furthermore, techniques like membrane potential-aware distillation and plug-and-play homeostatic spark have improved the training efficiency and stability of SNNs. Noteworthy papers include:

  • Exploring the Potentials of Spiking Neural Networks for Image Deraining, which proposes a novel Visual LIF neuron and achieves state-of-the-art performance with low energy consumption.
  • Temporal Dynamics Enhancer for Directly Trained Spiking Object Detectors, which introduces a spiking encoder and attention gating module to enhance temporal information modeling.
  • MD-SNN: Membrane Potential-aware Distillation on Quantized Spiking Neural Network, which leverages membrane potential knowledge distillation to mitigate accuracy degradation after quantization.

Sources

Revisiting Direct Encoding: Learnable Temporal Dynamics for Static Image Spiking Neural Networks

Exploring the Potentials of Spiking Neural Networks for Image Deraining

Temporal Dynamics Enhancer for Directly Trained Spiking Object Detectors

Efficient Eye-based Emotion Recognition via Neural Architecture Search of Time-to-First-Spike-Coded Spiking Neural Networks

Hybrid Temporal-8-Bit Spike Coding for Spiking Neural Network Surrogate Training

MD-SNN: Membrane Potential-aware Distillation on Quantized Spiking Neural Network

Plug-and-Play Homeostatic Spark: Zero-Cost Acceleration for SNN Training Across Paradigms

Built with on top of