Research map / Adversarial robustness and security / Attacks on text and graphs
DisGUIDE: Disagreement-Guided Data-Free Model Extraction
Jonathan Rosenthal, Eric Enouen, Hung Viet Pham, Lin Tan
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 |
|---|---|---|---|
| Train two clones and guide the generator with their disagreement loss | Data-free model extraction · CIFAR-10 soft-label | clone accuracy (%): 91.24 -> 94.02 ± 0.25 | Improved |
| Train two clones and guide the generator with their disagreement loss | Data-free model extraction · CIFAR-10 hard-label | clone accuracy (%): 84.51 -> 87.93 ± 1.74 | Improved |
| Train two clones and guide the generator with their disagreement loss | Data-free model extraction · CIFAR-100 hard-label | clone accuracy (%): 43.56 -> 62.04 ± 1.03 | Improved |
| Use DisGUIDE rather than DFME to reach the reported accuracy | CIFAR-10 soft-label extraction · DFME accuracy target | queries: 20M -> 3.13M ± 0.57M | Improved |
| Use DisGUIDE rather than DFMS to reach the reported accuracy | CIFAR-10 soft-label extraction · DFMS accuracy target | queries: 20M -> 7.01M ± 1.30M | Improved |
| Use DisGUIDE rather than DFMS to reach the reported accuracy | CIFAR-10 hard-label extraction · DFMS accuracy target | queries: 8M -> 4.02M ± 1.43M | Improved |
| Use DisGUIDE rather than DFMS to reach the reported accuracy | CIFAR-100 hard-label extraction · DFMS accuracy target | queries: 10M -> 3.87M ± 0.37M | Improved |
| Convert a fraction of generated images to grayscale (ablation) | DisGUIDE · CIFAR-10 soft-label | clone accuracy (%): 93.25 ± 0.15% -> 93.36 ± 0.16% | No clear effect |
| Convert a fraction of generated images to grayscale (ablation) | DisGUIDE · CIFAR-10 hard-label | clone accuracy (%): 85.54 ± 2.47% -> 87.93 ± 1.74% | Improved |
| Increase replay buffer size from 0.05M to 1M (ablation) | DisGUIDE · CIFAR-10 soft-label | clone accuracy (%): 93.06 -> 93.36 | No clear effect |
| Increase replay sampling frequency (ablation) | DisGUIDE · CIFAR-10 soft-label | clone accuracy (%): 92.3 -> 93.7 | Improved |
| Increase replay sampling frequency (ablation) | DisGUIDE · CIFAR-10 soft-label | runtime (hours): 7.0 -> 15.2 | Hurt |
Later papers that compared with this method
1 later accepted paper reports DisGUIDE as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- KOEnsAttack: Towards Efficient Data-Free Black-Box Adversarial Attacks via Knowledge-Orthogonalized Substitute Ensembles (ICCV 2025 · better on 2 of 2 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
- From Counterfactuals to Trees: Competitive Analysis of Model Extraction Attacks (NeurIPS 2025 · similarity 0.87)
- Exploring Query Efficient Data Generation Towards Data-Free Model Stealing in Hard Label Setting (AAAI 2025 · similarity 0.86)
- SAME: Sample Reconstruction against Model Extraction Attacks (AAAI 2024 · similarity 0.86)
- HoneypotNet: Backdoor Attacks Against Model Extraction (AAAI 2025 · similarity 0.85)
- Data-Free Hard-Label Robustness Stealing Attack (AAAI 2024 · similarity 0.85)
- Forget to Flourish: Leveraging Machine-Unlearning on Pretrained Language Models for Privacy Leakage (AAAI 2025 · similarity 0.83)
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.
