Research map / Adversarial robustness and security / Black-box and transfer attacks
Data-Free Model Extraction
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 |
|---|---|---|---|
| use the ℓ1 logit loss instead of KL divergence (ablation) | DFME · CIFAR10 | accuracy: 76.7% to 88.1% | Helped |
| use the ℓ1 logit loss instead of KL divergence (ablation) | DFME · SVHN | accuracy: 84.7% to 95.2% | Helped |
| recover logits by mean-correcting log-probabilities rather than using raw log-probabilities (ablation) | DFME · CIFAR10 | accuracy: 73.2% to 88.1% | Helped |
| recover logits by mean-correcting log-probabilities rather than using raw log-probabilities (ablation) | DFME · SVHN | accuracy: 94.4% to 95.2% | Helped |
| increase forward-difference directions from m=1 to m=10 (ablation) | DFME · CIFAR10 | queries to reach 85% accuracy: 10.04 to 20.00 | Hurt |
| randomly flip the m=1 approximate gradient with half probability (ablation) | DFME · CIFAR10 | accuracy: did not improve beyond 20% | Hurt |
| query the CIFAR10 victim with CIFAR100 rather than CIFAR10 | distillation-based extraction · CIFAR10 | extraction accuracy: 95.2% to 93.5% | Hurt |
| query the CIFAR10 victim with SVHN rather than CIFAR10 | distillation-based extraction · CIFAR10 | extraction accuracy: 95.2% to 66.6% | Hurt |
| query the CIFAR10 victim with MNIST rather than CIFAR10 | distillation-based extraction · CIFAR10 | extraction accuracy: 95.2% to 37.2% | Hurt |
| query the SVHN victim with random inputs rather than SVHN samples | distillation-based extraction · SVHN | extraction accuracy: 96.0% to 84.1% | Hurt |
| use DFME rather than the reproduced MAZE baseline | model extraction · CIFAR10 | accuracy: 45.6% to 88.1% | Helped |
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 0.87)
- HoneypotNet: Backdoor Attacks Against Model Extraction (AAAI 2025 · similarity 0.86)
- Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerable (NeurIPS 2024 · similarity 0.86)
- SAME: Sample Reconstruction against Model Extraction Attacks (AAAI 2024 · similarity 0.86)
- DataFreeShield: Defending Adversarial Attacks without Training Data (ICML 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.