Research map / Bandits and online learning / Contextual bandits and Thompson sampling
Online Corrupted User Detection and Regret Minimization
Zhiyong Wang, Jize Xie, Tong Yu, Shuai Li, John C.S. Lui
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 |
|---|---|---|---|
| OCCUD instead of GCUD detection (at 1M rounds) (ablation) | Synthetic · Synthetic | AUC: 0.502 -> 0.855 | Improved |
| OCCUD instead of GCUD detection (at 1M rounds) (ablation) | Movielens · Movielens | AUC: 0.492 -> 0.85 | Improved |
| OCCUD instead of GCUD detection (at 1M rounds) (ablation) | Amazon · Amazon | AUC: 0.518 -> 0.840 | Improved |
| OCCUD instead of GCUD detection (at 1M rounds) (ablation) | Yelp · Yelp | AUC: 0.510 -> 0.628 | Improved |
| more interaction rounds from 0.2M to 1M (ablation) | OCCUD, Synthetic · OCCUD, Synthetic | AUC: 0.599 -> 0.855 | Improved |
| more interaction rounds from 0.2M to 1M (ablation) | GCUD, Synthetic · GCUD, Synthetic | AUC: 0.477 -> 0.502 | No clear effect |
| more interaction rounds from 0.2M to 1M (ablation) | OCCUD, Yelp · OCCUD, Yelp | AUC: 0.452 -> 0.628 | Improved |
| at 0.2M rounds, OCCUD vs GCUD (ablation) | Yelp · Yelp | AUC: 0.473 -> 0.452 | Hurt |
| use robust weighted estimation with user clustering (RCLUB-WCU) instead of non-robust or single-vector baselines (ablation) | Synthetic, bandit regret · Synthetic, bandit regret | regret | Improved |
| lower corrupted-user ratio (3.5%) on Amazon (ablation) | RCLUB-WCU advantage over baselines · RCLUB-WCU advantage over baselines | regret | Hurt |
| per-user (Ind) estimators instead of one shared vector (ablation) | LinUCB / CW-OFUL · LinUCB / CW-OFUL | regret | Improved |
| corruption-aware CW-OFUL-Ind instead of LinUCB-Ind (ablation) | bandit baselines · bandit baselines | regret | Improved |
Closest papers by meaning
- Corruption-Robust Linear Bandits: Minimax Optimality and Gap-Dependent Misspecification (NeurIPS 2024 · similarity 0.84)
- Learning from Imperfect Human Feedback: A Tale from Corruption-Robust Dueling (ICLR 2025 · similarity 0.83)
- Online Linear Classification with Massart Noise (ICML 2025 · similarity 0.83)
- Locally Private and Robust Multi-Armed Bandits (NeurIPS 2024 · similarity 0.83)
- Pre-trained Behavioral Model for Malicious User Prediction on Social Platform (AAAI 2025 · similarity 0.83)
- Constructing Adversarial Examples for Vertical Federated Learning: Optimal Client Corruption through Multi-Armed Bandit (ICLR 2024 · similarity 0.82)
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.
