Reproduce “Hyb-KAN ViT: Hybrid Kolmogorov-Arnold Networks Augmented Vision Transformer” 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
Sainath Dey, Mitul Goswami, Jashika Sethi, Prasant Kumar Pattnaik
A hybrid architecture combines Kolmogorov-Arnold Networks with Vision Transformers through wavelet-based modules and spline functions, achieving 86.3% top-1 accuracy on ImageNet-1K while demonstrating improved parameter efficiency through Wav-KAN modules that enable 79.3% accuracy with only 6.9M parameters.
Key insights
- Wavelet-driven spectral priors improve segmentation performance
- Spline-based activation functions enhance detection tasks
- Hybrid architecture maintains steeper accuracy gains through 25 GFLOPs compared to standard ViTs
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.