Reproduce “Hardware Acceleration of Kolmogorov-Arnold Network (KAN) for Lightweight Edge Inference” 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
Wei-Hsing Huang, Jianwei Jia, Yuyao Kong, Faaiq Waqar, Tai-Hao Wen, Meng-Fan Chang, Shimeng Yu
A comprehensive hardware acceleration solution for Kolmogorov-Arnold Networks (KANs) is presented, integrating novel quantization, input generation, and weight mapping techniques with RRAM-based compute-in-memory (CIM) architectures. The work demonstrates significant improvements in energy consumption, area utilization, and computational latency for edge inference while maintaining high accuracy.
Key insights
- Kolmogorov-Arnold Networks (KANs) offer superior parameter efficiency compared to Deep Neural Networks (DNNs) for edge AI applications.
- Efficient hardware implementation of KANs on RRAM-based compute-in-memory (CIM) systems requires specialized quantization techniques, input generation methods, and weight mapping strategies to overcome B-spline complexity, memory access patterns, and non-ideal hardware effects.
- A holistic algorithm-hardware co-design approach, including automated hyperparameter optimization, can significantly improve energy, area, and latency efficiency while maintaining accuracy for KAN deployments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.