MicroSim
A binary code similarity detection system based on a contrastive siamese graph neural network that represents functions as IDA Pro Hex-Rays microcode graphs and performs similarity judgment through contrastive learning.
Timeline (last 90 days)
MicroSim is a binary code similarity detection system based on a contrastive Siamese GNN that represents each function as a graph built from IDA Pro Hex-Rays microcode.
The researchers evaluate MicroSim on the TREX cross-optimization benchmark, which contains 849,072 function-level microcode graphs covering optimization levels from -O0 to -O3.
MicroSim adopts a strict class-disjoint split, i.e., the training and test sets share no function keys.
Under the class-disjoint split protocol, MicroSim achieves a ROC-AUC of 0.9972 and an MCC of 0.9762 on the TREX benchmark.
MicroSim significantly outperforms baseline models on the TREX benchmark: Gemini (0.9332 ROC-AUC), SAFE (0.9752), jTrans (0.9740), and the Bag-of-Words baseline (0.9484).
All Facts (5)
MicroSim is a binary code similarity detection system based on a contrastive Siamese GNN that represents each function as a graph built from IDA Pro Hex-Rays microcode.
50%UnverifiedThe researchers evaluate MicroSim on the TREX cross-optimization benchmark, which contains 849,072 function-level microcode graphs covering optimization levels from -O0 to -O3.
50%UnverifiedMicroSim adopts a strict class-disjoint split, i.e., the training and test sets share no function keys.
50%UnverifiedUnder the class-disjoint split protocol, MicroSim achieves a ROC-AUC of 0.9972 and an MCC of 0.9762 on the TREX benchmark.
50%UnverifiedMicroSim significantly outperforms baseline models on the TREX benchmark: Gemini (0.9332 ROC-AUC), SAFE (0.9752), jTrans (0.9740), and the Bag-of-Words baseline (0.9484).
50%