Junjie Chen
AI + SE Seminar Series (February 28, 2025. 9 am – 10 am EST).
Comprehensive Effectiveness Enhancement of Code LLMs
Large language models for code (code LLMs), trained on extensive code corpora, have demonstrated substantial potential in software engineering applications. However, due to the complexity of practical software engineering tasks, existing code LLMs still face significant limitations in their effectiveness. Our work systematically enhances the effectiveness of code LLMs from three perspectives: (1) Data Augmentation, which improves training data distribution by constructing high-quality code datasets through augmentation techniques; (2) Data Denoising, which enhances deployed LLM effectiveness by mitigating noise in input code; and (3) Prompt Engineering, which optimizes LLM effectiveness in a plug-and-play manner by refining specification understanding. This talk provides a comprehensive introduction of these works, offering novel methodologies for improving the effectiveness of code LLMs in software engineering tasks.
Bio
Dr Junjie Chen is an Associate Professor in College of Intelligence and Computing, Tianjin University, China. He received his PhD from Peking University. His main research interests lie in software testing, SE for AI, and AI for SE. He has published 100+ research papers in top venues such as ICSE, FSE, ASE, and ISSTA, and has won five ACM SIGSOFT Distinguished Paper Awards and one ISSRE Best Research Paper Award. Also, he was ranked in World’s Top 2% of Scientists by Stanford University.
Visit: https://tjusail.github.io/people/chenjunjie/index.html
EECS Upcoming Events: https://lassonde.yorku.ca/eecs/eecs-upcoming-events/
Zoom Registration: Here