Research map / Optimization / Implicit bias and generalization
Optimization, Generalization and Differential Privacy Bounds for Gradient Descent on Kolmogorov–Arnold Networks
Puyu Wang, Junyu Zhou, Philipp Liznerski, Marius Kloft
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 |
|---|---|---|---|
| increase KAN width m from 8 to 512 | two-layer KAN · Multi-interval synthetic | empirical NTK margin: 0.0400 -> 0.1027 | Improved |
| use KAN instead of parameter-matched ReLU network | two-layer network · Multi-interval synthetic (m=512) | empirical NTK margin: 0.0197 -> 0.1027 | Improved |
| use KAN instead of parameter-matched ReLU network | two-layer network · XOR-gap synthetic (m=512) | empirical NTK margin: 0.0284 -> 0.0599 | Improved |
| use KAN instead of parameter-matched ReLU network | two-layer network · Checkerboard-gap synthetic (m=512) | empirical NTK margin: 0.0111 -> 0.0538 | Improved |
| use KAN instead of ReLU at small width | two-layer network · XOR-gap synthetic (m=8) | empirical NTK margin: 0.0226 -> 0.0243 | No clear effect |
| increase ReLU width from 64 to 512 | parameter-matched ReLU network · XOR-gap synthetic | empirical NTK margin: 0.0285 -> 0.0284 | No clear effect |
| increase width m beyond about 24 (GD) | two-layer KAN trained with GD · synthetic logistic data and MNIST | train and test accuracy | No clear effect |
| increase GD iterations T beyond about 256 | two-layer KAN trained with GD · synthetic logistic data and MNIST | test accuracy | No clear effect |
| increase width or iterations beyond about 64 under DP (epsilon=2) | two-layer KAN trained with DP-GD · synthetic logistic data | private test accuracy | Hurt |
Closest papers by meaning
- Generalization Bounds and Model Complexity for Kolmogorov–Arnold Networks (ICLR 2025 · similarity 0.90)
- Public-data Assisted Private Stochastic Optimization: Power and Limitations (NeurIPS 2024 · similarity 0.87)
- Differentially Private Stochastic Gradient Descent with Fixed-Size Minibatches: Tighter RDP Guarantees with or without Replacement (NeurIPS 2024 · similarity 0.87)
- How to Make the Gradients Small Privately: Improved Rates for Differentially Private Non-Convex Optimization (ICML 2024 · similarity 0.86)
- Implicit Bias of Gradient Descent for Non-Homogeneous Deep Networks (ICML 2025 · similarity 0.86)
- On the expressiveness and spectral bias of KANs (ICLR 2025 · similarity 0.86)
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.
