Yimeng Guo "Research on Software Quality Assurance: From Empirically-Grounded Knowledge to Neuro-Symbolic Approaches"

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

Title: Research on Software Quality Assurance: From Empirically-Grounded Knowledge to Neuro-Symbolic Approaches

Abstract:

In the AI era, dynamic languages such as Python underpin modern software systems, from data science to intelligent agents. While their flexibility accelerates development, it also introduces challenges in type safety, maintainability, and reliability. As systems grow in scale and complexity, stronger type-aware mechanisms become essential for effective software quality assurance.

In this talk, I present my research on type-aware quality assurance for Python, covering three aspects: type annotation generation to enrich static type information, type-check generation to ensure runtime correctness, and type-smell detection to uncover maintainability issues in evolving codebases. Although AI techniques such as large language models are powerful, they often lack reliability and guarantees in structured scenarios, and can incur high computational costs. To address these limitations, my work follows a neuro-symbolic perspective that combines neural techniques with program analysis and empirically-grounded knowledge from real-world software practice. This integration enables more accurate and efficient quality assurance across development, execution, and maintenance stages.

Building on this foundation, I will discuss future research ideas on type-safe AI4SE, focusing on neuro-symbolic approaches for reliable repository-level code generation and reasoning in agent-based systems, with stronger correctness guarantees and improved efficiency.