KQFuzz: LLM-Based Fuzzing for Quantum Libraries
A team of researchers has introduced KQFuzz, a knowledge-driven fuzzer designed specifically for quantum libraries, utilizing large language models (LLMs). This new approach overcomes the shortcomings of current LLM-based fuzzing techniques, which often suffer from limited flexibility and efficiency. KQFuzz integrates extensive codebase insights to enhance LLM-driven test creation, employing fitness-guided evaluations and dual-level mutations to navigate intricate execution paths and uncover possible bugs. Additionally, it presents an innovative prompting method customized for quantum applications, effectively using codebase knowledge to produce superior quantum seed programs. This research has been published on arXiv as a cross-type announcement.
Key facts
- KQFuzz is a knowledge-guided fuzzer for quantum libraries.
- It uses LLMs for test generation.
- It addresses limitations of existing LLM-based fuzzing approaches.
- It incorporates codebase knowledge for prompting.
- It uses fitness-guided evaluation and two-level mutations.
- It aims to explore complex execution paths and trigger bugs.
- The prompting scheme is tailored to quantum programs.
- The work is published on arXiv.
Entities
Institutions
- arXiv