Research map / Learning theory and algorithms / Prediction and decision losses
Prompt Certified Machine Unlearning with Randomized Gradient Smoothing and Quantization
Zijie Zhang, Yang Zhou, Xin Zhao, Tianshi Che, Lingjuan Lyu
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 |
|---|---|---|---|
| use PCMU instead of Fisher/NTK/SISA/DeltaGrad etc. for accuracy closeness to Retrain | PCMU · Fashion-MNIST and CIFAR-10 | accuracy: 64.33 vs 64.31 (Retrain) on CIFAR-10 10% | Improved |
| Replace baselines with PCMU, absolute accuracy difference to Retrain | PCMU · Fashion-MNIST | absolute accuracy difference: at least 5.56% | Improved |
| Replace baselines with PCMU, absolute accuracy difference to Retrain | PCMU · CIFAR-10 | absolute accuracy difference: at least 15.17% | Improved |
| Replace baselines with PCMU, absolute error difference to Retrain | PCMU · Fashion-MNIST | absolute error difference: at least 13.73% | Improved |
| Replace baselines with PCMU, absolute error difference to Retrain | PCMU · CIFAR-10 | absolute error difference: at least 16.01% | Improved |
| Use exact PCMU instead of variant PCMU-N (ablation) | PCMU · Fashion-MNIST and CIFAR-10 | accuracy and errors vs Retrain | Improved |
| Increase data removal ratio from 10% to 20% | PCMU · Fashion-MNIST | accuracy: 88.34 -> 88.34 | No clear effect |
| Increase data removal ratio from 10% to 20% | PCMU · CIFAR-10 | accuracy: 64.33 -> 64.33 | No clear effect |
| Vary data removal ratio from 5% to 20% | PCMU · Fashion-MNIST, CIFAR-10, SVHN | accuracy | No clear effect |
| Increase Gaussian smoothing sigma from 0.025 to 0.3 (ablation) | PCMU · Fashion-MNIST, CIFAR-10 | accuracy | Mixed |
| Use PCMU one-time unlearning instead of sequential DeltaGrad on runtime | PCMU · Fashion-MNIST | total runtime (s): 802 vs 694 | Hurt |
| Use PCMU instead of Fisher on runtime | PCMU · Fashion-MNIST 10% | total runtime (s): 2,686 -> 802 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeltaGrad — DeltaGrad: Rapid retraining of machine learning models (ICML 2020) · better on 4 of 4 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
- Certified Machine Unlearning via Noisy Stochastic Gradient Descent (NeurIPS 2024 · similarity 0.89)
- Certified Unlearning for Neural Networks (ICML 2025 · similarity 0.89)
- Hessian-Free Online Certified Unlearning (ICLR 2025 · similarity 0.89)
- Rewind-to-Delete: Certified Machine Unlearning for Nonconvex Functions (NeurIPS 2025 · similarity 0.89)
- Langevin Unlearning: A New Perspective of Noisy Gradient Descent for Machine Unlearning (NeurIPS 2024 · similarity 0.88)
- Machine Unlearning under Overparameterization (NeurIPS 2025 · similarity 0.88)
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.
