Research map / Learning theory and algorithms / Sample complexity
Lipschitz regularity of deep neural networks: analysis and efficient estimation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| split the bound computation across consecutive layers using SeqLip | AutoLip on MNIST-CNN · MNIST-CNN, 10 layers | Lipschitz upper-bound tightness relative to AutoLip: factor of 5 | Helped |
| use Greedy SeqLip with the 200 highest singular values of each linear layer | AlexNet Lipschitz estimation · AlexNet | Lipschitz upper bound: 3.62 \times 10^7 -> 5.45 \times 10^6 | Helped |
| use gradient-ascent Greedy SeqLip instead of exact SeqLip optimization (ablation) | SeqLip · synthetic MLP simulations | error relative to SeqLip: 0.01% | No clear effect |
| use gradient-ascent Greedy SeqLip when the exact optimum is computable (ablation) | SeqLip · settings where the exact optimum is computable | optimization error: at most 1% | No clear effect |
Closest papers by meaning
- ECLipsE: Efficient Compositional Lipschitz Constant Estimation for Deep Neural Networks (NeurIPS 2024 · similarity 0.93)
- Efficient and Accurate Estimation of Lipschitz Constants for Deep Neural Networks (NeurIPS 2019 · similarity 0.93)
- On Lipschitz Regularization of Convolutional Layers using Toeplitz Matrix Theory (AAAI 2021 · similarity 0.92)
- A Quantitative Geometric Approach to Neural-Network Smoothness (NeurIPS 2022 · similarity 0.92)
- Width Independent Bounds for the Local Lipschitz Constant of Deep Neural Networks at Random Initialization and after Lazy Training (ICML 2026 · similarity 0.91)
- Some Fundamental Aspects about Lipschitz Continuity of Neural Networks (ICLR 2024 · similarity 0.90)
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.