Research map / Adversarial robustness and security / Black-box and transfer attacks
DST: Dynamic Substitute Training for Data-Free Black-Box Attack
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 |
|---|---|---|---|
| Add GSIL to KL-based substitute training (ablation) | DST · Microsoft Azure example model, non-target, BIM | ASR: 70.28 -> 83.24 | Helped |
| Add DSSL to GSIL-based training (ablation) | DST · Microsoft Azure example model, non-target, BIM | ASR: 83.24 -> 95.02 | Helped |
| Replace MSE with KL divergence for matching substitute and target outputs (ablation) | DST · Microsoft Azure example model, non-target, BIM | ASR: 55.92 -> 70.28 | Helped |
| Add GSIL to KL-based substitute training (ablation) | DST · MNIST, non-target, probability-based | ASR: 48.99 -> 59.32 | Helped |
| Add DSSL to GSIL-based training (ablation) | DST · MNIST, non-target, probability-based | ASR: 59.32 -> 70.48 | Helped |
| Add DSSL to a ResNet-34 substitute with GSIL (ablation) | DST · CIFAR-100, target, probability-based | ASR: 21.48 -> 23.01 | Helped |
| Use DSSL to dynamically select substitute blocks rather than keeping a static ResNet-34 (ablation) | DST · MNIST, non-target, probability-based | ASR: 59.32 -> 70.48 | Helped |
| Use DSSL to dynamically select substitute blocks rather than keeping a static ResNet-34 (ablation) | DST · CIFAR-100, non-target, probability-based | ASR: 24.81 -> 31.39 | Helped |
| Add GSIL and DSSL instead of the KL-only baseline (ablation) | DST · Microsoft Azure example model, non-target, BIM | Train-Q: 30M -> 9M | Helped |
| Add GSIL to KL-based substitute training (ablation) | DST · Microsoft Azure example model, non-target, BIM | Train-Q: 30M -> 13M | Helped |
| Use DST rather than DDG+AST | Data-free black-box attack · Microsoft Azure example model, target, label-based | ASR: 71.48 -> 75.90 | Helped |
Closest papers by meaning
- AGS: Affordable and Generalizable Substitute Training for Transferable Adversarial Attack (AAAI 2024 · similarity 0.87)
- Data-Free Universal Attack by Exploiting the Intrinsic Vulnerability of Deep Models (AAAI 2025 · similarity 0.84)
- Power of Diversity: Enhancing Data-Free Black-Box Attack with Domain-Augmented Learning (AAAI 2025 · similarity 0.83)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.83)
- Small Scale Data-Free Knowledge Distillation (CVPR 2024 · similarity 0.83)
- Out of Thin Air: Exploring Data-Free Adversarial Robustness Distillation (AAAI 2024 · similarity 0.82)
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.