N2NMatcher: New AI framework makes binary code matching 2x more resilient to inlining
A hierarchical graph neural network finds stable function anchors to beat compiler optimizations...
N2NMatcher is a new framework for program-level binary code similarity analysis (BCSA) designed to stay resilient to function inlining. It learns to predict stable “anchor” functions using a hierarchical graph neural network that encodes ACFG-FCG representations, then performs anchor-bounded decomposition and matches the resulting modules with learned graph embeddings. Experiments show it improves decomposition quality and module matching accuracy over existing methods, enabling more effective program-level BCSA.
- Uses hierarchical GNN over ACFG-FCG graphs to predict stable anchor functions
- Anchor-bounded decomposition avoids breakage from function inlining
- Improves module matching accuracy vs. existing BCSA approaches in benchmarks
Why It Matters
More reliable binary similarity detection means better vulnerability hunting and malware analysis in real-world, compiler-optimized binaries.