|
I am an undergraduate student from GSAI, Renmin University of China. I am fortunate to be advised by Zhicheng Dou, working on Natural Language Processing and Information Retrieval. I also work closely with Heng Ji from UIUC and Weiyan Shi from Stanford University. I had a fantastic experience as an exchange student at UC Berkeley. I'm looking for a Ph.D. position starting from Fall 2024. Feel free to contact me if you are interested!
Resume /
Research Statement
|
|
|
|
I work on Augmented Language Models. I have done several works regarding (1) general interaction, e.g., interaction benchmark, (2) the cross application of language models and information retrieval (IR) systems, e.g., retrieval augmented LMs and LM-based IR. I'm currently working on a boarder range of LM Augmentation: controlling interactive language models. I'm also interested in AI alignment topics like scalable oversight. |
|
Xingyao Wang*, Zihan Wang*, Jiateng Liu, Yangyi Chen, Lifan Yuan, Hao Peng, Heng Ji International Conference on Learning Representations (ICLR) 2024 [paper] [code] [website] Current LLM evaluations focus on single-turn and overlook multi-turn real-world scenarios. We introduce the MINT benchmark to assess LLMs in interaction with tools and language feedback. Our study of 20 LLMs shows they benefit from multi-turn interactions, but current RLHF and SIFT methods might hinder this. MINT aims to encourage research on LLM multi-turn capabilities, especially in open-source models. |
|
Zihan Wang, Yujia Zhou, Yiteng Tu, Zhicheng Dou. Conference on Information and Knowledge Management (CIKM) 2023, Oral Presentation [paper] [code] We propose Neural Optimized Vocabularial doc-ids for model-based information retrieval. NOVO doc-ids consist of non-overlapping n-gram sets to identify documents, optimized through query denoising modeling and retrieval tasks. We further demonstrate how N-grams lead to interpretability in understanding documents and queries. |
|
Jiongnan Liu, Jiajie Jin, Zihan Wang, Jiehan Cheng, Zhicheng Dou, Ji-Rong Wen Arxiv Preprint [paper] [code] We develop a RETreival-Augmented LLM toolkit. It provides more plug-and-play modules to support better interaction between IR systems and LLMs, including request rewriting, document retrieval, passage extraction, answer generation, and fact checking modules. |
|
Zihan Wang, Hongjin Qian, Zhicheng Dou China National Conference on Computational Linguistics (CCL) 2023 [paper][code][poster] We propose a self-supervised learning method on structured documents for conditional question answering, consisting of a conditional question generation approach and a contrastive learning objective. We propose a pipeline able to generate multiple answers with detailed conditions. |
|
|
|
Website design from Jon Barron |