Research map / Adversarial robustness and security / Black-box and transfer attacks
ME-Net: Towards Effective Adversarial Robustness with Matrix Estimation
Yuzhe Yang, Guo Zhang, Dina Katabi, Zhi Xu
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 matrix-estimation preprocessing | ResNet-18 on CIFAR-10 · transfer-based black-box CW attack | adversarial accuracy: 8.9% -> 93.6% | Improved |
| add matrix-estimation preprocessing | ResNet-18 on CIFAR-10 · transfer-based black-box FGSM attack | adversarial accuracy: 24.8% -> 92.2% | Improved |
| add matrix-estimation preprocessing to adversarial training | ResNet-18 on CIFAR-10 · white-box BPDA attack, 1000 steps | adversarial accuracy: 45.0% -> 52.8% | Improved |
| add matrix-estimation preprocessing to adversarial training | WideResNet on CIFAR-10 · white-box BPDA attack, 1000 steps | adversarial accuracy: 46.8% -> 55.1% | Improved |
| add matrix-estimation preprocessing | ResNet-18 on CIFAR-10 · clean test | accuracy: 93.4% -> 94.9% | Improved |
| add matrix-estimation preprocessing | LeNet on MNIST · clean test | accuracy: 98.8% -> 99.2% | Improved |
| add matrix-estimation preprocessing | ResNet-18 on SVHN · clean test | accuracy: 95.0% -> 96.0% | Improved |
| add matrix-estimation preprocessing | DenseNet-121 on Tiny-ImageNet · clean test, Top-1 | accuracy: 66.4% -> 67.7% | Improved |
| replace nuclear norm minimization with USVT (ablation) | ME-Net on CIFAR-10 · black-box transfer PGD, 40 steps | adversarial accuracy: 91.0% -> 89.4% | No clear effect |
| replace nuclear norm minimization with Soft-Impute (ablation) | ME-Net on CIFAR-10 · black-box transfer PGD, 40 steps | adversarial accuracy: 91.0% -> 91.3% | No clear effect |
| use observing probability p: 0.4 to 0.6 instead of p: 0.8 to 1 (ablation) | pure ME-Net on CIFAR-10 · white-box BPDA attack, 1000 steps | adversarial accuracy: 23.5% -> 41.6% | Improved |
| use observing probability p: 0.4 to 0.6 instead of p: 0.8 to 1 (ablation) | ME-Net with adversarial training on CIFAR-10 · white-box BPDA attack, 1000 steps | adversarial accuracy: 47.4% -> 52.8% | Improved |
Later papers that compared with this method
1 later accepted paper reports ME-Net as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- API-Net: Robust Generative Classifier via a Single Discriminator (ECCV 2020 · better on 3 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Achieving Adversarial Robustness by Enforcing Feature Consistency Across Bit Planes (CVPR 2020 · similarity 0.88)
- One Man's Trash Is Another Man's Treasure: Resisting Adversarial Examples by Adversarial Examples (CVPR 2020 · similarity 0.87)
- A Direct Approach to Robust Deep Learning Using Adversarial Networks (ICLR 2019 · similarity 0.87)
- Countering Adversarial Images using Input Transformations (ICLR 2018 · similarity 0.87)
- Detecting and Diagnosing Adversarial Images with Class-Conditional Capsule Reconstructions (ICLR 2020 · similarity 0.87)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 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.
