Zichuan Liu

Zichuan Liu | 刘子川

I am a master student at Nanjing University, supervised by Prof. Chunlin Chen, and I am privileged to work closely with Dr. Wei Cheng, A/Prof. Dongsheng Luo, and A/Prof. Mengnan Du. I have also spent time at Microsoft Research, advised by Dr. Lei Song and managed by Dr. Jiang Bian. Before that, I was a research intern at DAMO Academy of Alibaba Group, advised by Dr. Qingsong Wen. Previously, I completed my B.S. in Computer Science at Wuhan University of Technology, working with A/Prof. Rui Zhang.

My research interests focus on Reinforcement Learning, eXplainable Artificial Intelligence, and Trustworthy LLMs, especially with applications in multiple sequences, offline models, and dynamic environments, allowing them to explicitly adapt to non-stationary task distributions in real-world. For your information, please see my research statement.

I will pause my research from May to Oct. If have any issues, please email me.

Email / CV / Google Scholar / Github / Twitter / Linkedin

Invited Talk

"Perturbation-based Techniques for Explaining Sequence Predictions", April, 2024. See This Slide.

"Interpretable and Efficient Multi-Agent Reinforcement Learning", February, 2024. See This Slide.

Research

Representative papers are highlighted, * denotes contribution equally.

	Protecting Your LLMs with Information Bottleneck Zichuan Liu, Zefan Wang, Linjie Xu, Jinyu Wang, Lei Song, Tianchun Wang, Chunlin Chen, Wei Cheng, Jiang Bian Neural Information Processing Systems (NeurIPS), Vancouver, Canada, 2024 *(CCF-A, CORE A)** project page / paper / code / slides / bibtex Our LLM protector provides an effective defence against harmful prompts without losing much of its information!
	Knowing What Not to Do: Leverage Language Model Insights for Action Space Pruning in Multi-agent Reinforcement Learning Zhihao Liu, Xianliang Yang, Zichuan Liu, Yifan Xia, Wei Jiang, Yuanyu Zhang, Lijuan Li, Guoliang Fan, Lei Song, Jiang Bian arXiv:2405.16854, 2024 (Under Review) paper / code / bibtex Learning exploration functions to prune multi-agent action spaces by the LLM.
	TimeX++: Learning Time-Series Explanations with Information Bottleneck Zichuan Liu, Tianchun Wang, Jimeng Shi, Xu Zheng, Zhuomin Chen, Lei Song, Wenqian Dong, Jayantha Obeysekera, Farhad Shirani, Dongsheng Luo International Conference on Machine Learning (ICML), Vienna, Austria, 2024 *(CCF-A, CORE A)** paper / code / bibtex Introduce information theory in explaining time-series.
	Position: Rethinking Post-Hoc Search-Based Neural Approaches for Solving Large-Scale Traveling Salesman Problems Yifan Xia, Xianliang Yang, Zichuan Liu, Zhihao Liu, Lei Song, Jiang Bian International Conference on Machine Learning (ICML), Vienna, Austria, 2024 *(Oral Paper, Top 2%, CCF-A, CORE A)** paper / code / bibtex Doubt about the effectiveness of ML-based heatmap generation in Traveling Salesman Problems.
	Explaining Time Series via Contrastive and Locally Sparse Perturbations Zichuan Liu, Yingying Zhang, Tianchun Wang, Zefan Wang, Dongsheng Luo, Mengnan Du, Min Wu, Yi Wang, Chunlin Chen, Lunting Fan, Qingsong Wen International Conference on Learning Representations (ICLR), Vienna, Austria, 2024 *(CORE A)** project page / paper / code / slides / bibtex A novel perturbation-based method for time series explanation.
	Higher Replay Ratio Empowers Sample-Efficient Multi-Agent Reinforcement Learning Linjie Xu, Zichuan Liu, Alexander Dockhorn, Diego Perez-Liebana, Jinyu Wang, Lei Song, Jiang Bian IEEE Conference on Games (IEEE CoG), 2024 paper / code / bibtex We look into the training mechanism of MARL and increase replay ratios to enhance sample efficiency.
	RCAgent: Cloud Root Cause Analysis by Autonomous Agents with Tool-Augmented Large Language Models Zefan Wang, Zichuan Liu, Yingying Zhang, Aoxiao Zhong, Lunting Fan, Lingfei Wu, Qingsong Wen Conference on Information and Knowledge Management (CIKM), Boise, USA, 2024 *(CCF-B, CORE A)** paper / code / bibtex We introduce RCAgent, a tool-augmented LLM autonomous agent for cloud root cause analysis.
	NA²Q: Neural Attention Additive Model for Interpretable Multi-Agent Q-Learing Zichuan Liu, Yuanyang Zhu, Chunlin Chen International Conference on Machine Learning (ICML), Honolulu, USA, 2023 *(CCF-A, CORE A)** paper / code / presentation / bibtex We present a novel method providing inherent intelligibility of collaboration behavior on multi-agent reinforcement learning.
	Boosting Value Decomposition via Unit-Wise Attentive State Representation for Cooperative Multi-Agent Reinforcement Learning Qingpeng Zhao, Yuanyang Zhu, Zichuan Liu, Zhi Wang, Chunlin Chen arXiv:2305.07182, 2023 (Under Review) paper / code / bibtex A transformer framework to significantly improve the performance of multi-agent collaboration.
	MIXRTs: Toward Interpretable Multi-Agent Reinforcement Learning via Mixing Recurrent Soft Decision Trees Zichuan Liu, Yuanyang Zhu, Zhi Wang, Yang Gao, Chunlin Chen arXiv:2209.07225, 2022 (Under Review) paper / code / bibtex We propose a novel interpretable architecture that can represent explicit decision processes via the root-to-leaf path of recurrent soft decision trees.
	Spatial-temporal Conv-sequence Learning with Accident Encoding for Traffic Flow Prediction Zichuan Liu, Rui Zhang, Chen Wang, Zhu Xiao, Hongbo Jiang IEEE Transactions on Network Science and Engineering (TNSE), 2022 (SCI, JCR-Q1, IF=6.6) paper / code / bibtex STCL improves traffic prediction by capturing short-term temporal dependencies and accident impact through convolution and sequence learning.
	Multi View Spatial-Temporal Model for Travel Time Estimation Zichuan Liu, Zhaoyang Wu, Meng Wang, Rui Zhang International Conference on Advances in Geographic Information Systems (SIGSPATIAL), Beijing, China, 2021 (CCF-C, CORE A) paper / code / bibtex This solution achieved high performance on large-scale taxi trajectory data and won a runner-up award at SIGSPATIAL'2021 GISCUP.

Experience

	Microsoft Research Asia 2023.11 - 2024.05 Research Intern at Machine Learning Group Milestones: ICML'24a, ICML'24b, ArXiv'24a, ArXiv'24b, CoG'24 Research Advisor: Dr. Lei Song and Manager: Dr. Jiang Bian
	Alibaba DAMO Academy & Alibaba Cloud 2023.06 - 2023.11 Research Intern Milestones: ICLR'24, CIKM'24 Research Advisor: Dr. Qingsong Wen
	Nanjing University 2022.09 - Present M.S. in Electronic Engineering Milestones: ICML'23, ArXiv'22, ArXiv'23 GPA: 89.38/100, Supervisor: Prof. Chunlin Chen
	Wuhan University of Technology 2018.09 - 2022.06 B.S. in Computer Science Milestones: IEEE TNSE, SIGSPATIAL'21 GPA: 91.10/100, Advisor: A/Prof. Rui Zhang