Peng Yin Choo

Traditional fuzzing struggles with efficiency, as maximizing code coverage does not guarantee the discovery of additional vulnerabilities. To solve this, the study introduces topology-aware node evaluation (TANE-Pool), a deep learning model that proactively predicts defects to guide the fuzzing process. The model analyzes the structural properties of a program’s attributed control flow graph (ACFG) via a diffusion attention mechanism. This process identifies fragile code regions and generates a static vulnerability score (SVS) for each basic block. The fuzzer then uses this score to prioritize test cases, concentrating its efforts on the areas most likely to contain flaws. Evaluated on the Juliet test suite and a set of real-world programs, TANE Pool demonstrates superior prediction accuracy. Its integration into a fuzzer significantly enhances the rate of vulnerability discovery, proving that a defect-prediction-guided approach is a more efficient and effective strategy for software security testing.