Siqi Gu "From Test Generation to Trustworthy Repair: Verification and Feedback for LLM-Based Software Engineering"

This talk can be viewed via zoom. (Note: this talk will be recorded)

Abstract
Large language models can generate software tests, but ensuring that these tests execute correctly and detect faults remains challenging. This talk presents my doctoral research on improving LLM-based unit test generation through verification and feedback. I will introduce TestART, which combines test generation with template-based repair; SUMMIT, which integrates LLM-generated candidates into search-based testing; and AttnProbe, which uses coverage-guided context pruning and mutation feedback to improve fault detection. I will also discuss recent findings on skill selection and evolution in testing agents. Building on this work, I will outline opportunities to support trustworthy automated program repair through targeted test generation, patch correctness assessment, and validated reuse of agent experience.

Bio
Siqi Gu recently defended her Ph.D. in Software Engineering at Nanjing University, supervised by Prof. Zhenyu Chen. During her doctoral studies, she spent one year at University College London working with Prof. Earl T. Barr. Her research focuses on LLM-based software testing, search-based test generation, and verification and feedback mechanisms for software engineering agents. Her work has appeared in ACM TOSEM, ISSTA, and IEEE QRS.