Undergrad student, National Taiwan University
1 paper at NeurIPS 2025
We combine LLM-synthesized performance-characterizing constraints with fuzzing to uncover difficult-to-find code inefficiencies and generate performance-stressing tests.