NDSS 2019 Send Hardest Problems My Way: Probabilistic Path Prioritization for Hybrid Fuzzing

Hybrid fuzzing combines fuzzing's efficiency with concolic execution's precision, but existing approaches like demand launch and optimal switch strategies suffer from inefficiencies. The discriminative dispatch strategy addresses these limitations by using a Monte Carlo-based probabilistic path prioritization model that quantifies path difficulty through sampling-based probability calculations. This lightweight model identifies and prioritizes only the most difficult paths for concolic execution, avoiding the overhead of processing all inputs. Evaluation on CQE and LAVA datasets demonstrates that this approach outperforms state-of-the-art systems like Driller, achieving higher code coverage (18.9% vs. 3.8%) and discovering more vulnerabilities (12 vs. 5) while maintaining efficient resource utilization.