XUE Boyang
Beyond Hsueh
XUE Boyang
Beyond Hsueh
I am a fourth-year Ph.D. candidate of System Engineering and Engineering Management Department, The Chinese University of Hong Kong, fortunately supervised by
Prof. Wong Kam-Fai.
My research interests focus on natural language processing, trustworthy AI, Bayesian learning, etc.
I'm positively seeking for research internship and exchange opportunities of the laboratoties in universities or companies.
Expect collaborations in LLM related research to exchange great ideas and welcome to contact me.
⭐️ Reliable LLM/Agent: LLMs are prone to generate certain responses to all questions, even though for those out of LLMs’ capabilities and thus resulting in hallucinations. Therefore, we propose to improve the reliability of LLMs, where LLMs can correctly and confidently answer their known questions, express uncertainty to questions they are not sure, and refuse to answer unknown questions, to avoid misleading users. We employ a framework for reliability alignment by first enabling LLMs to perceive their knowledge or ability boundaries, then aligning the perceptions with factual expressions, and finally generalizing to unseen tasks. [GitHub]
⭐️ Bayesian Learning: Large-scale neural language models with highly complex architectures and point-estimated parameters fail to account for model uncertainty and is prone to overfitting and poor generalization on low-resource tasks given limited training data. We apply Bayesian learning framework to account for uncertainty modeling to LMs. Different types of uncertainty modeling can be applied to model parameters, model architectures, and hidden representations with efficient variational inference approaches to estimate the latent posterior distribution.
⭐️ Multilingual LLM: In multilingual communication, individuals typically draw upon the knowledge bases inherent to their native languages. Yet, the cultural divergences across languages often pose obstacles to effective interaction. To address this challenge, we focus on transferring culture-specific knowledge advantages embedded in each language to others in alignment, aiming to “Rebuild the Babel Tower”.
Visiting Researcher in Web Intelligence Group, UCL AI Centre.
Ph.D. Student in MoE Key Lab of High Confidence Software Technologies.
Research Assistant and Ph.D. Student in Human-Computer Communications Lab.
Research Intern in Intelligent Manufacturing and Data Science Lab.
Team Leader in Intelligent Application and Robotics Team.
My favourite musicians: Claude Debussy, Frederic Chopin, Ryuichi Sakamoto, Hisaishi Jean, et al. I love romantic classical music best. I'm preparing ABRSM Exam.
Served as a part-time author writing academic blogs about NLP/ML/LLM and comments on some classic books and movies at Zhihu and WeChat Officials.
My favorite singers: Jay Chou, JJ Lin, Eason Chan, Leslie Cheung Kwok Wing, et al. I like various vocalization skills like xi qiang, bel canto, and whistle register.
Served as a contract photographer at Vision China 500px Community.