Reproduce “Temporal-wise Attention Spiking Neural Networks for Event Streams Classification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Man Yao, Huanhuan Gao, Guangshe Zhao, Dingheng Wang, Yihan Lin, Zhaoxu Yang, Guoqi Li
Researchers from Xi’an Jiaotong University and Tsinghua University developed Temporal-wise Attention Spiking Neural Networks (TA-SNNs) to efficiently classify event streams by dynamically assessing frame importance. This approach achieves state-of-the-art accuracy on multiple event-based datasets while significantly reducing computational costs through intelligent input frame pruning.
Key insights
- Temporal attention, when applied at depth layers within a Spiking Neural Network, consistently improves classification accuracy on event stream data.
- The proposed Input Attention Pruning mechanism can intelligently discard up to 50% of input frames at inference, leading to substantial computational savings without compromising accuracy, and sometimes even improving it.
- Advanced deep learning techniques like temporal attention can be effectively integrated into Spiking Neural Networks, narrowing the performance gap with ANNs while retaining SNNs' inherent energy efficiency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.