Posts by Collection

portfolio

publications

Fusing Reward and Dueling Feedback in Stochastic Bandits

Published in ICML, 2025

This paper proposes novel algorithms (ElimFusion and DecoFusion) to fuse absolute reward and relative dueling feedback in multi-armed bandits, achieving regret bounds that adaptively leverage the more informative feedback type. Theoretical and empirical results demonstrate significant performance gains over baselines.

Recommended citation: Wang, X. et al. (2025). "Fusing Reward and Dueling Feedback in Stochastic Bandits." arXiv preprint arXiv:2504.15812.
Download Paper

Practical Adversarial Attacks on Stochastic Bandits via Fake Data Injection

Published in arXiv, 2025

We propose Fake Data Injection, a practical attack on stochastic bandits where the adversary injects limited, bounded fake feedback. Our strategies efficiently deceive UCB and Thompson Sampling into favoring a target arm with minimal cost, exposing critical vulnerabilities in real-world applications.

Recommended citation: Zeng, Q. et al. (2025). "Practical Adversarial Attacks on Stochastic Bandits via Fake Data Injection." arXiv preprint arXiv:2505.21938.
Download Paper