CovRL: Fuzzing JavaScript Engines with Coverage-Guided Reinforcement Learning for LLM-based Mutation
원문 초록
Fuzzing is an effective bug-finding technique but it struggles with complex systems like JavaScript engines that demand precise grammatical input. Recently, researchers have adopted language models for context-aware mutation in fuzzing to address this problem. However, existing techniques are limited in utilizing coverage guidance for fuzzing, which is rather performed in a black-box manner. This paper presents a novel technique called CovRL (Coverage-guided Reinforcement Learning) that combines Large Language Models (LLMs) with reinforcement learning from coverage feedback. Our fuzzer, CovRL-Fuzz, integrates coverage feedback directly into the LLM by leveraging the Term Frequency-Inverse Document Frequency (TF-IDF) method to construct a weighted coverage map. This map is key in calculating the fuzzing reward, which is then applied to the LLM-based mutator through reinforcement learning. CovRL-Fuzz, through this approach, enables the generation of test cases that are more likely to discover new coverage areas, thus improving vulnerability detection while minimizing syntax and semantic errors, all without needing extra post-processing. Our evaluation results indicate that CovRL-Fuzz outperforms the state-of-the-art fuzzers in terms of code coverage and bug-finding capabilities: CovRL-Fuzz identified 48 real-world security-related bugs in the latest JavaScript engines, including 39 previously unknown vulnerabilities and 11 CVEs.
함께 살펴볼 논문
공통 태그를 근거로 연결했습니다. 실제 인용 관계나 논문의 우수성을 뜻하지 않습니다.
평가에서 최적화로: Python CWE 예측을 위한 계층 인식 학습 신호
From Evaluation to Optimisation: Hierarchy-Aware Training Signals for CWE Prediction in Python
원래 ALPHA 벤치마크는 Python에서 CWE 수준 취약점 예측을 평가하기 위한 분류체계 인식 페널티를 도입했으며, 이 페널티가 이론적으로 학습 신호로도 사용될 수 있다고 제안했다. 본 논문은 그 검증을 제공한다. 우리는 세 가지 전달 메커니즘을 비교한다: 지도 미세조정, 이중 헤드 분류 손실, 그리고 정규화된 페널티에서 유도된 밀집 보상을 사용한 강화학습. 분포 이동 하에서 지도 방식은 일관되게 제로샷 기준선 아래로 퇴화하는 반면, GRPO는 성공한다. 최적 정책은 탐욕 디코딩에서 SVEN(Security Hardening and Adversarial Testing) 데이터셋의 Qwen2.5-Coder-7B 누적 ALPHA 페널티를 27.9% 줄이고, 샘플링 디코딩에서는 25.5% 줄인다(p = 0.005, Welch's t-test). 4.5배 더 큰 제로샷 교사 모델과 통계적 동등성에 도달한다. 계층적 페널티를 학습 신호로서의 가치는 크게 그 전달 방식의 직접성에 달려 있다고 결론짓는다.
Reasoning on a Budget: Miniaturizing DeepSeek R1 with SFT-GRPO Alignment for Instruction-Tuned LLMs
Meta Reinforcement Learning for Multi-Task Offloading in Vehicular Edge Computing
Investigating the Performance and Reliability, of the Q-Learning Algorithm in Various Unknown Environments
Margin Trader: A Reinforcement Learning Framework for Portfolio Management with Margin and Constraints
Catch: Collaborative Feature Set Search for Automated Feature Engineering
토론
이 논문과 연결된 포럼 글이 아직 없습니다.