Publications
Accepted
Offline Reward Perturbation Boosts Distributional Shift in Online RL
40th Conference on Uncertainty in Artificial Intelligence (UAI), 2024
Proposes a stealthy data poisoning attack for offline-to-online reinforcement learning that preserves offline performance while inducing harmful distribution shift during online fine-tuning.
Poisoning Generative Replay in Continual Learning to Promote Forgetting
40th International Conference on Machine Learning (ICML), 2023
Studies poisoning attacks against generative replay in continual learning, showing how poisoned generators can promote forgetting while maintaining strong performance on the current task.
Data-Independent Memory Hard Functions: New Attacks and Stronger Constructions
39th Annual International Cryptology Conference (CRYPTO), 2019
Analyzes the security of data-independent memory-hard functions, introducing new attacks and stronger constructions with improved resistance to known pebbling attacks.
Preprints / Under Review
Adaptive Data Harvesting for Universal-Constraint Learning
Under review, IJCAI 2026
Formulates collocation selection as a sequential decision-making problem and learns an adaptive sampling policy that improves convergence, stability, and constraint satisfaction in universal-constraint learning settings such as Lyapunov neural networks and physics-informed neural networks.