The field of Spiking Neural Networks (SNNs) is moving towards improving their efficiency, scalability, and adaptability to various applications. Recent developments focus on leveraging SNNs' potential for low power consumption and event-driven processing, making them suitable for edge devices and real-time processing. Researchers are exploring innovative training methods, such as ANN-SNN distillation and hybrid block-wise replacement, to enhance SNN learning and accuracy. Moreover, there is a growing interest in designing hardware accelerators for SNNs, which can efficiently process spike-form data and reduce energy consumption. Another notable trend is the development of novel architectures and algorithms for SNNs, such as those enabling temporal flexibility, shortest path finding, and place disambiguation. Noteworthy papers include Efficient ANN-Guided Distillation, which proposes a block-wise replacement strategy for ANN-guided SNN learning, and LightSNN, which presents a rapid and efficient Neural Network Architecture Search technique specifically tailored for SNNs. Region Masking to Accelerate Video Processing on Neuromorphic Hardware is also notable for its region masking strategy to reduce computation and data movement for events arising from unimportant regions. These advancements demonstrate the potential of SNNs to revolutionize the field of artificial intelligence and neuromorphic computing.
Advances in Spiking Neural Networks
Sources
Efficient ANN-Guided Distillation: Aligning Rate-based Features of Spiking Neural Networks through Hybrid Block-wise Replacement
Region Masking to Accelerate Video Processing on Neuromorphic Hardware
Replay4NCL: An Efficient Memory Replay-based Methodology for Neuromorphic Continual Learning in Embedded AI Systems
A Digital Machine Learning Algorithm Simulating Spiking Neural Network CoLaNET
Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness
Hardware Efficient Accelerator for Spiking Transformer With Reconfigurable Parallel Time Step Computing
VESTA: A Versatile SNN-Based Transformer Accelerator with Unified PEs for Multiple Computational Layers
Input-Triggered Hardware Trojan Attack on Spiking Neural Networks
Threshold Adaptation in Spiking Networks Enables Shortest Path Finding and Place Disambiguation
LightSNN: Lightweight Architecture Search for Sparse and Accurate Spiking Neural Networks