FDSE v2: Variable Importance Guided Hybrid Fuzzing (Competition Contribution)

Abstract is a hybrid fuzzing tool that automatically generates high-coverage test suites for C programs. In prior work, we introduced an initial version of that used fuzzing-based pre-analysis to enhance symbolic execution. This paper presents , featuring a novel constraint tree component that more effectively coordinates the fuzzer and concolic execution engine, thereby reducing ineffective search in fuzzing. The constraint tree models the space of path constraints already explored by concolic execution and enables fine-grained partitioning of input variables based on their semantic relevance. During seed mutation, the fuzzer focuses only on the most important variable groups identified through this analysis. Concolic execution excels at navigating complex path conditions involving arithmetic or logical guards, while the fuzzer leverages the resulting high-quality seeds as starting points for deeper exploration under extended time budgets. By reducing redundant efforts between the two engines, achieves significant improvements in both code coverage and vulnerability detection effectiveness.