Siteng Kang

Siteng Kang

PhD Candidate in Computer Science

Publications

Offline Reward Perturbation Boosts Distributional Shift in Online RL

Published in 40th Conference on Uncertainty in Artificial Intelligence (UAI), 2024

We proposed a data poisoning attack on offline to online reinforcement learning to stealthily promote distribution shift.

Recommended citation: Yu, Z.*, Kang, S.*, & Zhang, X. (*equal contribution). (2024, July). Offline Reward Perturbation Boosts Distributional Shift in Online RL. In The 40th Conference on Uncertainty in Artificial Intelligence. https://openreview.net/pdf?id=wbwTF909Ve

Poisoning Generative Replay in Continual Learning to Promote Forgetting

Published in 40th International Conference on Machine Learning (ICML), 2023

We proposed a data poisoning attack on generative models in the context of replay based continual learning.

Recommended citation: Kang, S., Shi, Z., & Zhang, X. (2023, July). Poisoning generative replay in continual learning to promote forgetting. In International Conference on Machine Learning (pp. 15769-15785). PMLR. https://proceedings.mlr.press/v202/kang23c/kang23c.pdf

Data-Independent Memory Hard Functions: New Attacks and Stronger Constructions

Published in 39th Annual International Cryptology Conference (CRYPTO), 2019

We proposed a Data-Independent Memory Hard Functions and showed that it provides the best resistance to known pebbling attacks

Recommended citation: Blocki, J., Harsha, B., Kang, S., Lee, S., Xing, L., & Zhou, S. (2019, August). Data-independent memory hard functions: New attacks and stronger constructions. In Annual International Cryptology Conference (pp. 573-607). Cham: Springer International Publishing. https://eprint.iacr.org/2018/944.pdf