Research map / Architectures and efficiency / CNN design
Lookup multivariate Kolmogorov-Arnold Networks
Sergey Pozdnyakov, Philippe Schwaller
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use 2D multivariate functions (lmKAN) instead of 1D FastKAN | lmKAN · CIFAR-10 fully connected | accuracy | Improved |
| increase grid resolution G beyond 12 for lmKAN (ablation) | lmKAN hidden_dim 256 · R32->R1 function approximation | MSE | Hurt |
| increase grid resolution to very fine for FastKAN (ablation) | FastKAN · CIFAR-10 fully connected | accuracy | Hurt |
| add Hessian regularization to lmKAN (ablation) | lmKAN · methane Distances Polynomials | test RMSE | Improved |
| replace MLP with lmKAN | MLP backbone · general function approximation | inference FLOPs at matched MSE: up to 6× | Improved |
| replace MLP with lmKAN | MLP backbone · methane configurations | H100 wall-clock time at matched accuracy: more than 10× | Improved |
| replace MLP convs with lmKAN convs | CNN · CIFAR-10 | inference FLOPs at matched accuracy: 1.6-2.1× | Improved |
| replace MLP convs with lmKAN convs | CNN · ImageNet-1k | inference FLOPs at matched accuracy: 1.7× | Improved |
| reduce tile 16x16 to 8x8 (allowing G=40) (ablation) | lmKAN CUDA kernel · kernel benchmark on H100 | slowdown vs linear: 8× -> 9.5× slower, but 27.5× -> 88.5× better per-parameter efficiency | Mixed |
Closest papers by meaning
- KAN: Kolmogorov–Arnold Networks (ICLR 2025 · similarity 0.86)
- Generalization Bounds and Model Complexity for Kolmogorov–Arnold Networks (ICLR 2025 · similarity 0.85)
- On the expressiveness and spectral bias of KANs (ICLR 2025 · similarity 0.85)
- Incorporating Arbitrary Matrix Group Equivariance into KANs (ICML 2025 · similarity 0.84)
- PowerMLP: An Efficient Version of KAN (AAAI 2025 · similarity 0.84)
- Kolmogorov-Arnold Transformer (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
