Reproduce “HiFuzz: Hierarchical Reinforcement Learning for Semantic-Aware and Adaptive CPU Fuzzing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ya Wang, Hanwei Fan, Zhenguo Liu, Xiaofeng Zhou, Yangdi Lyu, Jiang Xu, Wei Zhang
HiFuzz presents a hierarchical reinforcement learning framework for CPU fuzzing, employing a dual-agent architecture, semantic-aware intrinsic rewards, and an adaptive coverage mechanism. It achieved up to a 48.9% improvement in control register coverage on Rocket Core and detected more bugs than baseline fuzzers on complex RISC-V architectures like BOOM.
Key insights
- Decomposing the complex program generation task into a hierarchical RL structure (Program Agent for macro, Basic Block Agent for micro) effectively manages the vast action space and maintains structural integrity.
- Providing dense, semantic-aware intrinsic feedback at the basic block level significantly mitigates the sparse reward problem inherent in hardware fuzzing, guiding agents without immediate hardware simulation.
- Dynamically rebalancing exploration across hardware modules is crucial to overcome the 'masking effect' of reward bias, ensuring comprehensive coverage and discovery of deeply embedded bugs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.