Research map / Adversarial robustness and security / Backdoor and poisoning
ReTrace: Reinforcement Learning-Guided Reconstruction Attacks on Machine Unlearning
Mengyao Ma, Shuofeng Liu, Jason Xue, Surya Nepal, Guangdong Bai
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 |
|---|---|---|---|
| move from black-box to white-box access (adding loss and gradient traces) (ablation) | ReTrace · CIFAR-100 (approximate unlearning) | SR: 55.9% -> 73.1% | Improved |
| add grey-box loss-level traces to black-box access (ablation) | ReTrace · CIFAR-100 distribution-level | FID: 142.1 -> 135.0 | Improved |
| use white-box access instead of black-box access (ablation) | ReTrace · CIFAR-100 distribution-level (exact) | KL: 3.62 -> 2.49 | Improved |
| grey-box vs black-box on Food-101 exact unlearning (ablation) | ReTrace · Food-101 (exact unlearning) | SR: 48.9% -> 54.7% -> 50.4% | Mixed |
| approximate instead of exact unlearning target | ReTrace · CIFAR-100 (white-box) | SR: 62.4% -> 73.1% | Improved |
| use ReTrace instead of UIA | UIA · CIFAR-100 (exact unlearning, white-box) | MSE: 0.35 -> 0.20 | Improved |
| use ReTrace instead of HRec | HRec · CIFAR-100 (approximate unlearning, white-box) | SR: 43.0% -> 73.1% | Improved |
| change unlearned class (Aquarium fish, Bed, Bridge) (ablation) | ReTrace · CIFAR-100 classes (white-box) | SR: 59.1%, 60.5%, 60.8% (exact) | No clear effect |
| increase access level from black-box to white-box on text (ablation) | ReTrace · AG News (text) | BLEU: 2.8 -> 4.1 | Improved |
| increase access level on text (Business MMD) (ablation) | ReTrace · AG News Business | MMD: 0.46 -> 0.32 | Improved |
Closest papers by meaning
- Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerable (NeurIPS 2024 · similarity 0.89)
- The Unseen Threat: Residual Knowledge in Machine Unlearning under Perturbed Samples (NeurIPS 2025 · similarity 0.88)
- From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularization (NeurIPS 2025 · similarity 0.87)
- Learning to Rewind via Iterative Prediction of Past Weights for Practical Unlearning (AAAI 2025 · similarity 0.87)
- Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyond (ICML 2025 · similarity 0.87)
- Adversarial Mixup Unlearning (ICLR 2025 · similarity 0.87)
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.
