CodeExtract: Enhancing Binary Code Similarity Detection with Code Extraction Techniques
summary
CodeExtract implements proactive inlining for single-call functions plus duplicate basic-block extraction to turn inlined code back into calls, operating on assembly-level disassembly and control-flow graphs. It applies define-use based register renaming and basic-block similarity checks (only extract fragments with ≥3 blocks) to normalize functions before feeding LB-BCSD models (evaluated including ASMVector). Function inlining cuts model accuracy from ~80% to ~50% (30% drop); CodeExtract recovers ~20% absolute accuracy (inline emulation gives ~10%), and finds >92% of inline blocks are simple and recoverable.
AI-written summary · about
chapters (7)
- 0:00CodeExtract intro
- 3:24Function inlining impact
- 5:19Proactive inlining
- 7:27Code extraction solution
- 9:14Duplicate code challenges
- 11:51Evaluation results
- 13:14Q&A
related talks
30mEnhancing CTI Processes with Code Search Technology
23mDeduBB: Binary Code Size Reduction via Post-Link Basic Block Deduplication
25mEnhancing Malware Code Similarity Detection through Vectorsearch and TLSH
27mBinSim: Trace-based Semantic Binary Diffing...
57mCode Genome - Fingerprinting Code to Build Trustworthy SBOMs
24m(Ab)using Compiler Tools
27mA Study of Inline Assembly in Solidity Smart Contracts
22mA Scalable, Ensemble Approach for Building and Visualizing Deep Code-Sharing Networks
