Research map / Adversarial robustness and security / Certified and empirical robustness
Expressive 1-Lipschitz Neural Networks for Robust Multiple Graph Learning against Adversarial Attacks
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 |
|---|---|---|---|
| Replace GroupSort-style fixed 1-Lipschitz layers with ERNN (Weibull activation, graph-adaptive Kl, norm-constrained weights) | graph classification robustness · BZR, BZR MD, MUTAG under 5% perturbed edges | Accuracy: ERNN highest on all three datasets (e.g., MUTAG 0.788 vs GCN-SVD 0.653) | Helped |
| Use ERNN instead of baselines for graph matching under RND, NEA, GMA attacks | graph matching · AS, CAIDA, DBLP with varying perturbed edges | Hits@1: ERNN highest in most tests, Hits@1 > 0.249 on AS | Helped |
| Use Weibull activation with graph-adaptive Kl in place of ReLU (ablation) | ERNN graph matching · AS, CAIDA, DBLP with 5% perturbed edges | Hits@1: ERNN-R (ReLU) lower than full ERNN in ablation | Helped |
| Fix Kl=1 instead of graph-adaptive Kl (ablation) | ERNN graph matching · AS, CAIDA, DBLP with 5% perturbed edges | Hits@1 | Hurt |
| Remove Lipschitz constraint (regular network ERNN-N) vs ERNN-R with Lipschitz-bounded architecture (ablation) | ERNN graph matching · AS, CAIDA, DBLP with 5% perturbed edges | Hits@1 | Hurt |
| Increase Kl from 1 to larger values (ablation) | ERNN graph matching · AS, CAIDA, DBLP under RND, NEA, GMA attacks | Hits@1 | Mixed |
| Use Kl-Lipschitz Weibull activation instead of ReLU, Leaky ReLU, Sigmoid, tanh (ablation) | ERNN graph matching · Graph matching Hits@1 across attacks | Hits@1 | Helped |
| Use Weibull-based 1-Lipschitz ERNN instead of GroupSort/BCOP on image classification | image classification · MNIST and CIFAR-10 depth 3 | Accuracy: MNIST 0.97 vs 0.94 (BCOP); CIFAR-10 0.55 vs 0.48 (GroupSort) | Helped |
| Impose 1-Lipschitz constraint (ERNN) vs unbounded network | graph matching Lipschitz constant · AS, CAIDA, DBLP | Lipschitz constant: 0.910 vs 2880 on AS | Helped |
Closest papers by meaning
- Lipschitz Bounds and Provably Robust Training by Laplacian Smoothing (NeurIPS 2020 · similarity 0.89)
- A Recipe for Improved Certifiable Robustness (ICLR 2024 · similarity 0.87)
- ECLipsE: Efficient Compositional Lipschitz Constant Estimation for Deep Neural Networks (NeurIPS 2024 · similarity 0.87)
- Robust Graph Neural Networks via Unbiased Aggregation (NeurIPS 2024 · similarity 0.86)
- Approximation theory for 1-Lipschitz ResNets (NeurIPS 2025 · similarity 0.86)
- A Simple and Yet Fairly Effective Defense for Graph Neural Networks (AAAI 2024 · 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.