Research map / Adversarial robustness and security / Certified and empirical robustness
Robust Design of Deep Neural Networks Against Adversarial Attacks Based on Lyapunov Theory
Check 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 |
|---|---|---|---|
| Apply per-layer Lyapunov spectral-norm regularization | 3-layer forward-net · MNIST, PGD ε=0.1–0.3 | mean output change: 0.244 ≤ 0.435; 0.490 ≤ 0.615; 0.736 ≤ 0.753 vs 1.233; 2.446; 3.641 | Helped |
| Apply per-layer Lyapunov spectral-norm regularization | ResNet50 · ImageNet, FGM ε=0.1–0.4 | adversarial test accuracy: 0.61, 0.51, 0.44, 0.38 vs 0.57, 0.45, 0.40, 0.34 | Helped |
| Use per-layer Lyapunov spectral regularization instead of ℓ2 regularization | ResNet50 · ImageNet, FGM ε=0.1–0.4 | adversarial test accuracy: 0.61, 0.51, 0.44, 0.38 vs 0.58, 0.46, 0.39, 0.32 | Helped |
| Increase global Lyapunov parameters | AlexNet on CIFAR10 · CIFAR10, PGD and FGM attacks | test accuracy under attack | Helped |
| Use Leaky ReLU rather than ReLU | DNN layers · Adversarial robustness | robustness | Helped |
| Add a residual skip connection | ResNet building block · Adversarial robustness | robustness | Hurt |
| Apply a single ℓ2 weight-decay regularizer | DNNs under attack · CIFAR10, PGD and FGM attacks | test accuracy under attack | No clear effect |
Closest papers by meaning
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.91)
- Exploring Architectural Ingredients of Adversarially Robust Deep Neural Networks (NeurIPS 2021 · similarity 0.90)
- Adversarial Robustness in Graph Neural Networks: A Hamiltonian Approach (NeurIPS 2023 · similarity 0.90)
- Towards the first adversarially robust neural network model on MNIST (ICLR 2019 · similarity 0.90)
- Lyapunov-Stable Deep Equilibrium Models (AAAI 2024 · similarity 0.89)
- Stability and Generalization of Adversarial Training for Shallow Neural Networks with Smooth Activation (NeurIPS 2024 · similarity 0.89)
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.