Research map / Architectures and efficiency / Neural architecture search
Anti-Bandit Neural Architecture Search for Model Defense
Hanlin Chen, Baochang Zhang, Song Xue, Xuan Gong, Hong Liu, Rongrong Ji, David Doermann
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 LCB-based fair evaluation and UCB pruning (anti-bandit) instead of UCB-only selection (ablation) | ABanditNAS · MNIST PGD-100 | accuracy (%): 95.4 -> 95.7 | Improved |
| add LCB fair evaluation on top of UCB pruning (ablation) | UCBNAS (pruning) · MNIST PGD-100 | accuracy (%): 94.96 -> 95.7 | Improved |
| add anti-bandit pruning to UCBNAS (ablation) | UCBNAS · MNIST search cost | GPU days: 0.13 -> 0.08 | Improved |
| add LCB fair evaluation on top of UCB pruning (ablation) | UCBNAS (pruning) · CIFAR-10 PGD-20 | accuracy (%): 45.33 -> 50.51 | Improved |
| add LCB fair evaluation on top of UCB pruning (ablation) | UCBNAS (pruning) · CIFAR-10 PGD-7 | accuracy (%): 54.55 -> 50.0 | Hurt |
| train 10 cells instead of 6 | ABanditNAS · CIFAR-10 PGD-7 | accuracy (%): 50.0 -> 58.74 | Improved |
| widen network to 100 initial channels for 6 cells | ABanditNAS-6 · CIFAR-10 PGD-7 | accuracy (%): 50.0 -> 51.24 | Mixed |
| use ResNet-18 instead of a network copy to generate black-box attacks | ABanditNAS-6 · CIFAR-10 black-box PGD | accuracy (%): 76.13 -> 78.0 | Improved |
| set lambda to 0.7 (ablation) | ABanditNAS · hyperparameter lambda sweep | robustness | Improved |
Closest papers by meaning
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.89)
- Towards Accurate and Robust Architectures via Neural Architecture Search (CVPR 2024 · similarity 0.88)
- TRNAS: A Training-Free Robust Neural Architecture Search (ICCV 2025 · similarity 0.88)
- Insights from the Use of Previously Unseen Neural Architecture Search Datasets (CVPR 2024 · similarity 0.85)
- Training-free Neural Architecture Search through Variance of Knowledge of Deep Network Weights (CVPR 2025 · similarity 0.85)
- Robustifying and Boosting Training-Free Neural Architecture Search (ICLR 2024 · similarity 0.85)
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.
