Research map / Adversarial robustness and security / Certified and empirical robustness
LipNeXt: Scaling up Lipschitz-based Certified Robustness to Billion-parameter Models
Kai Hu, Haoqi Hu, Matt Fredrikson
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 |
|---|---|---|---|
| add periodic polar retraction (ablation) | LipNeXt L32W2048 · CIFAR-10 | CRA at 36/255: 72.3 -> 73.2 | Improved |
| add Lookahead (ablation) | LipNeXt L32W2048 · CIFAR-10 | CRA at 36/255: 72.4 -> 73.2 | Improved |
| add spatial shift on 1/16 of channels (vs none) (ablation) | LipNeXt L32W2048 · CIFAR-10 | CRA at 36/255: 53.6 -> 73.2 | Improved |
| shift 1/16 of channels instead of all channels (ablation) | LipNeXt L32W2048 · CIFAR-10 | CRA at 36/255: 67.4 -> 73.2 | Improved |
| add learnable positional embedding with circular padding (ablation) | LipNeXt L32W2048 · CIFAR-10 | CRA at 36/255: 72.1 -> 73.2 | Improved |
| zero padding instead of circular padding, with positional embedding (ablation) | LipNeXt L32W2048 · CIFAR-10 | CRA at 36/255: 73.2 -> 72.6 | Hurt |
| use beta-Abs (beta=0.75) instead of MinMax (ablation) | LipNeXt L32W2048 · CIFAR-10 | CRA at 36/255: 72.7 -> 73.2 | Improved |
| beta=1.0 instead of 0.75 (ablation) | LipNeXt L32W2048 · CIFAR-10 | CRA at 36/255: 73.2 -> 72.5 | Hurt |
| increase width from 1024 to 4096 | LipNeXt depth 32 · ImageNet-400 classes | CRA at eps=1: 22.9 -> 30.0 | Improved |
| increase depth from 8 to 64 | LipNeXt width 2048 · ImageNet-400 classes | CRA at eps=1: 22.4 -> 26.9 | Improved |
| increase depth from 64 to 128 | LipNeXt width 2048 · ImageNet-400 classes | CRA at eps=1: 26.9 -> 26.8 | No clear effect |
| add diffusion-generated synthetic data and scale to L32W2896 | LipNeXt · CIFAR-10 | CRA at 36/255: 73.2 -> 81.7 | Improved |
Closest papers by meaning
- Improved techniques for deterministic l2 robustness (NeurIPS 2022 · similarity 0.90)
- A Recipe for Improved Certifiable Robustness (ICLR 2024 · similarity 0.88)
- ECLipsE: Efficient Compositional Lipschitz Constant Estimation for Deep Neural Networks (NeurIPS 2024 · similarity 0.87)
- Enhancing Certified Robustness via Block Reflector Orthogonal Layers and Logit Annealing Loss (ICML 2025 · similarity 0.87)
- Certified Robustness Under Bounded Levenshtein Distance (ICLR 2025 · similarity 0.86)
- Efficient Robust Conformal Prediction via Lipschitz-Bounded Networks (ICML 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.
