Researchers have developed HiFuzz, a novel framework utilizing hierarchical reinforcement learning to improve CPU fuzzing efficiency. This system employs a two-layer generation process, with a Program Agent managing global layout and a Basic Block Agent handling instruction filling. To address reward sparsity, HiFuzz incorporates an adaptive coverage reward mechanism and a semantic-aware basic block encoder for intrinsic feedback. Evaluations on RISC-V cores show HiFuzz surpasses current state-of-the-art fuzzers in both coverage and bug detection. AI
IMPACT Enhances CPU verification efficiency and bug detection capabilities through advanced AI techniques.
RANK_REASON The cluster contains a research paper detailing a new methodology for CPU fuzzing. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- Basic Block Agent
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- Program Agent
- reinforcement learning
- RISC-V
- ScienceCast
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
