Research map / Learning theory and algorithms / Prediction and decision losses
UltraRE: Enhancing RecEraser for Recommendation Unlearning via Error Decomposition
Read the paperCheck 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 Balanced k-means division with Optimal Balanced Clustering (OBC) (ablation) | UltraRE stage I · ML-100K/ML-1M/ADM | inertia: 46.32% | Helped |
| replace Balanced Random Division with OBC (ablation) | UltraRE stage I · ML-100K/ML-1M/ADM | inertia: 22.01% | Helped |
| use OBC instead of unbalanced k-means (ablation) | UltraRE stage I · ML-100K/ML-1M/ADM | inertia | No clear effect |
| Balanced k-means vs Balanced Random Division (ablation) | RecEraser stage I · ML-100K/ML-1M/ADM | inertia | Hurt |
| use average combiner instead of logistic regression combiner (ablation) | UltraRE stage III · DMF | NDCG@10: 3.53% | Hurt |
| use average combiner instead of logistic regression combiner (ablation) | UltraRE stage III · LightGCN | NDCG@10: 6.51% | Hurt |
| replace attention combiner with logistic regression combiner (ablation) | UltraRE stage III · ML-100K/ML-1M/ADM | NDCG@10 | No clear effect |
| UltraRE vs RecEraser (model utility) | UltraRE · ML-100K | NDCG@10/HR@10: 1.31% | Helped |
| UltraRE vs RecEraser (model utility) | UltraRE · ML-1M | NDCG@10/HR@10: 1.26% | Helped |
| optimal balanced clustering replaces BKM in stage I (efficiency) | UltraRE · ML-100K/ML-1M/ADM | stage I time: 98.11% | Helped |
| simplified LR combiner in stage III (efficiency) | UltraRE · ML-100K/ML-1M/ADM | stage III time: 11.95% | Helped |
| ensemble retraining (SISA/RecEraser/UltraRE) vs full Retrain | ensemble frameworks · ML-100K/ML-1M/ADM | total running time: 85.39% | Helped |
Closest papers by meaning
- CURE4Rec: A Benchmark for Recommendation Unlearning with Deeper Influence (NeurIPS 2024 · similarity 0.92)
- Multi-Modal Recommendation Unlearning for Legal, Licensing, and Modality Constraints (AAAI 2025 · similarity 0.87)
- RRL: Recommendation Reverse Learning (AAAI 2024 · similarity 0.86)
- Exploring Criteria of Loss Reweighting to Enhance LLM Unlearning (ICML 2025 · similarity 0.86)
- Erase Then Rectify: A Training-Free Parameter Editing Approach for Cost-Effective Graph Unlearning (AAAI 2025 · similarity 0.86)
- Leveraging Per-Instance Privacy for Machine Unlearning (ICML 2025 · similarity 0.85)
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.