Research map / Adversarial robustness and security / Attacks on text and graphs
Gradient-Free Adversarial Training Against Image Corruption for Learning-based Steering
Yu Shen, Laura Yu Zheng, Manli Shu, Weizi Li, Tom Goldstein, Ming Lin
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 frequency-space branch (ablation) | Ours (Nvidia CNN) · SullyChen single perturbation | mCE: 49.47 -> 37.30 | Improved |
| add frequency-space branch (ablation) | Ours (Nvidia CNN) · SullyChen unseen perturbation | mCE: 76.20 -> 61.61 | Improved |
| add frequency-space branch (ablation) | Ours (Nvidia CNN) · SullyChen combined perturbation | MMAI: 33.24 -> 33.15 | No clear effect |
| add frequency-space branch (ablation) | Ours (Nvidia CNN) · SullyChen clean | AMAI: 0.93 -> 2.12 | Improved |
| use per-factor worst-case dataset selection with two-stage training instead of simple data augmentation | Nvidia CNN · SullyChen single perturbation | mCE: 51.34 -> 37.30 | Improved |
| use proposed method instead of AugMix | Nvidia CNN · SullyChen clean | AMAI: -5.23 -> 2.12 | Improved |
| use proposed method instead of AugMix | Nvidia CNN · Audi single perturbation | mCE: 55.27 -> 10.27 | Improved |
| use proposed method instead of AugMix | Nvidia CNN · Waymo combined perturbation | MMAI: 22.95 -> 21.34 | Hurt |
| use proposed method instead of AugMix | ResNet152 (Honda) · Honda clean | AMAI: -4.23 -> -0.96 | Mixed |
| use proposed method instead of AugMix on CIFAR-100-C | 4 CNN architectures · CIFAR-100-C | mean corruption error (%): 38.3 -> 22.0 | Improved |
| use proposed method instead of MaxUp worst-case-only selection | Nvidia CNN · SullyChen single perturbation | mCE: 66.56 -> 37.30 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MaxUp — MaxUp: Lightweight Adversarial Training With Data Augmentation Improves Neural Network Training (CVPR 2021) · better on 4 of 4 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
- PhysGAN: Generating Physical-World-Resilient Adversarial Examples for Autonomous Driving (CVPR 2020 · similarity 0.86)
- AdvDO: Realistic Adversarial Attacks for Trajectory Prediction (ECCV 2022 · similarity 0.85)
- Testing DNN-based Autonomous Driving Systems under Critical Environmental Conditions (ICML 2021 · similarity 0.85)
- Steerable Adversarial Scenario Generation through Test-Time Preference Alignment (ICLR 2026 · similarity 0.85)
- SlowTrack: Increasing the Latency of Camera-Based Perception in Autonomous Driving Using Adversarial Examples (AAAI 2024 · similarity 0.85)
- NeuRAD: Neural Rendering for Autonomous Driving (CVPR 2024 · similarity 0.84)
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.
