Weichao Zhou
I am a Postdoc at MIT. My research interests span reinforcement learning (RL) and formal methods. My research has focused on incorporating data-driven methods into safety-critical autonomous systems. I have also worked on robustness analysis of neural network control systems, backdoor attack detection in neural networks across various applications, including computer vision and reinforcement learning, and building robust information retrieval systems using large language models (LLMs).
Here is my research statement.
Contact: zhouw534@mit.edu
Updates
[2024/09/25] Our submission “Rethink Inverse Reinforcement Learning: from Data Alignment to Task Alignment" has been accepted to NeurIPS 2024 !!!
[2024/09/22] Our submission “HyQE: Ranking Contexts with Hypothetical Query Embeddings", has been accepted to EMNLP 2024 Findings !!!
[2024/02/21] Our submission “Temporal Logic Specification-Conditioned Decision Transformer for Offline Safe Reinforcement Learning" has been accepted to ICML 2024!!!
[2023/09/28] New preprint of “PAGAR: Taming Reward Misalignment in Inverse Reinforcement Learning-Based Imitation Learning with Protagonist Antagonist Guided Adversarial Reward” is available. [code]
[2023/010/20] Our workshop paper “Universal Trojan Signatures in Reinforcement Learning” has been accepted to NeurIPS 2023 BUGS Workshop !!
[2023/010/20] Our submission “POLAR-Express: Efficient and Precise Formal Reachability Analysis of Neural-Network Controlled Systems” has been accepted to TCAD 2023 !!
Select Papers
