Xiaocheng Yang

I am currently a PhD student in Data Science at College of William & Mary under the supervision of Professor Qingyun Wang. I earned my master's degree in Computer Science at University of Illinois at Urbana-Champaign (UIUC) supervised by Professor Gokhan Tur and Professor Dilek Hakkani-Tür, and my bachelor's degree in Computer Science with Mathematics minor at New York University Shanghai supervised by Professor Yik-Cheung (Wilson) Tam.

At New York University Shanghai, I received Dean's Undergraduate Research Fund (DURF) and conducted research on neural symbolic methods for boosting large language models' reasoning ability and investigated the advantage of using Prolog langauge (a logic programming langauge) as the generation output in terms of data augmentation for finetunining. After that, I explored how large language models could be used to solve dialogua state tracking, which is a key component in task-oriented dialogues, and witnessed a major improvement compared with previous methods.

At University of Illinois at Urbana-Champaign, I joined ConvAI Lab and worked on conversational AI and agentic systems. I explored how benign friction in conversation helps conversational AI better assist users. After that, I was particularly interested in multi-agent systems (MAS). I participated in benchmarking MAS covering a comprehensive range of tasks and also proposed a novel method called Gradient-Based Connections (GBC) to optimize the prompts of multi-agent systems, which significantly improved the performance of the system in conversational scenarios. Meanwhile, I participated in CELaRAI project, where I developed a system to help early literacy education for children using large language models.