NESA Framework
A program-analysis framework that combines deterministic and uncertain evidence.
NESA is a program-analysis framework that integrates deterministic and uncertain information.
NESA incorporates uncertain signals—including variable names, information obtained through testing, and feedback from large language models—into deterministic program-analysis methods to improve analysis precision.
My work primarily involved expanding the use of large language models’ understanding of programs as uncertain evidence and evaluating its effectiveness in data-race analysis. By comparing hard and soft evidence, I demonstrated the effectiveness of soft evidence.