UROP Proceedings 2022-23

School of Science Department of Mathematics 53 Research in AI and Machine Learning Supervisor: ZHANG, Tong / MATH Student: WANG, Yipeng / MAEC Course: UROP1100, Summer The UROP project investigates the potential of prompt engineering to conduct textual style transfer in large language models with limited training corpus. During the UROP course, I have become familiar with conducting chatbot experiments using LMFlow, an application under development by the HKUST machine learning research group. To gain an adequate competency for my role, I have also learnt a substantial amount of knowledge in areas including Linux, bash scripts, git, prompt learning and many more. By the end of the program, I managed to create and finetune a styled chatbot of Li Bai, Shakespeare (with large corpus) and made a starting progress with the fictional character chatbot Fischel (with very limited corpus). Besides, I am currently designing an algorithm that aim to optimize the evaluation prompt for the chatbot. The report is a comprehensive documentation of my learning journey, with reflections on my progress and contributions. Research in AI and Machine Learning Supervisor: ZHANG, Tong / MATH Student: WONG, Ho Nam / COSC Course: UROP1100, Spring UROP2100, Summer Instead of only talking about what I have done in this course, this report rather elaborates on my journey in these two months. First I would talk about the work I have accomplished in the Neurips paper after submission, and I would continue participating in the rebuttal phase as well. Then I would talk about my understanding on OOD generalization. Then I would talk about my journey on learning theory and optimization such as papers I have read, workshops I have attended, and some other opportunities I have grasped. In the end, I would talk about my future plan on research.