Research map / Learning theory and algorithms / Prediction and decision losses
Efficient and Accurate Estimation of Lipschitz Constants for Deep Neural Networks
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 |
|---|---|---|---|
| use LipSDP-Neuron instead of CPLip and SeqLip | five-hidden-layer MNIST networks · MNIST | Lipschitz upper bound (lower is tighter) | Helped |
| use LipSDP-Neuron (neuron-level multipliers) instead of LipSDP-Layer (one multiplier per layer) (ablation) | LipSDP · one-hidden-layer, 3000 units | computation time (s): 473.25 -> 734.63 | Hurt |
| add cross-neuron coupling constraints (LipSDP-Network) over LipSDP-Neuron (ablation) | LipSDP · one-hidden-layer, 100 neurons | Lipschitz bound tightness | Helped |
| split the network into small sub-networks and multiply bounds | LipSDP · 100-layer and 500-layer random networks | scalability: 50,000 neurons: about 12 min (Neuron), about 4 min (Layer) | Helped |
| train with LP-Train or PGD-Train instead of standard Adam | MNIST classifiers · MNIST | Lipschitz bound | Helped |
| replace ReLU with leaky ReLU activation | three-hidden-layer MNIST networks · MNIST | Lipschitz constant | Hurt |
Closest papers by meaning
- ECLipsE: Efficient Compositional Lipschitz Constant Estimation for Deep Neural Networks (NeurIPS 2024 · similarity 0.95)
- Width Independent Bounds for the Local Lipschitz Constant of Deep Neural Networks at Random Initialization and after Lazy Training (ICML 2026 · similarity 0.93)
- Some Fundamental Aspects about Lipschitz Continuity of Neural Networks (ICLR 2024 · similarity 0.91)
- DP-SGD Without Clipping: The Lipschitz Neural Network Way (ICLR 2024 · similarity 0.90)
- Novel Quadratic Constraints for Extending LipSDP beyond Slope-Restricted Activations (ICLR 2024 · similarity 0.90)
- A Recipe for Improved Certifiable Robustness (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.