Reproduce “SechKAN: Kolmogorov-Arnold Networks with Hyperbolic Secant Functions” 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

Hoang-Thang Ta

SechKAN pairs localized sech-basis responses at grid centers with a learned projection that compresses them to one value per feature before feature mixing. In a parameter-matched CIFAR-10 ablation, it reaches 51.07% validation accuracy versus 39.83% without projection. On one Shallow Water trajectory, its test relative L2 error is 0.29879, lower than FasterKAN’s; this does not establish generalization across PDE conditions or faster computation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.