Research map / Learning theory and algorithms / Prediction and decision losses
Deep Unlearning via Randomized Conditionally Independent Hessians
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 |
|---|---|---|---|
| Replace conditional mutual information with L-CODEC for raw-data Markov Blanket identification | L-CODEC + CIT · 3D-Bullseye raw data | TPR: 0.75 -> 1.00 | Helped |
| Replace conditional mutual information with L-CODEC for raw-data Markov Blanket identification | L-CODEC + CIT · 3D-Bullseye raw data | Time (s): 5124.22 -> 402.10 | Helped |
| Replace conditional mutual information with L-CODEC for feature-map Markov Blanket identification | L-CODEC + CIT · 3D-Bullseye feature maps | TPR: 0.875 -> 0.75 | Hurt |
| Replace conditional mutual information with L-CODEC for feature-map Markov Blanket identification | L-CODEC + CIT · 3D-Bullseye feature maps | Time (s): 516.19 -> 117.29 | Helped |
| Use L-FOCI instead of CIT for feature-map Markov Blanket selection | L-CODEC + L-FOCI · 3D-Bullseye feature maps | Time (s): 516.19 -> 0.464 | Helped |
| Use L-FOCI instead of CIT for feature-map Markov Blanket selection | L-CODEC + L-FOCI · 3D-Bullseye feature maps | FPR: 0.00 -> 0.50 | Hurt |
| Use a same-size random parameter subset instead of the L-FOCI-selected subset (ablation) | L-FOCI unlearning · MNIST logistic regression | sample gradient norm after removal | Hurt |
| Use K-FAC Hessian approximation instead of L-CODEC-selected Hessian blocks (ablation) | MNIST multilayer perceptron unlearning · MNIST | validation metrics after unlearning | Hurt |
| Increase weight decay / L2 regularization (ablation) | L-FOCI unlearning and retraining · MNIST | validation performance under repeated removals | Helped |
| Increase the privacy parameter epsilon | L-FOCI unlearning · Market-1501 person re-identification | number of supported removals: 0.0005; 0.1 | Helped |
Closest papers by meaning
- Unified Gradient-Based Machine Unlearning with Remain Geometry Enhancement (NeurIPS 2024 · similarity 0.88)
- Towards Source-Free Machine Unlearning (CVPR 2025 · similarity 0.88)
- Hessian-Free Online Certified Unlearning (ICLR 2025 · similarity 0.87)
- Machine Unlearning under Overparameterization (NeurIPS 2025 · similarity 0.87)
- Partially Blinded Unlearning: Class Unlearning for Deep Networks from Bayesian Perspective (AAAI 2025 · similarity 0.87)
- Learning to Unlearn: Instance-Wise Unlearning for Pre-trained Classifiers (AAAI 2024 · 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.