REVDECODE: Enhancing Binary Function Matching with Context-Aware Graph Representations and Relevance Decoding
Tongwei Ren
34th USENIX Security Symposium (USENIX Security '25) · Day 3 · Software Security 4: Fuzzing and Other Software Analysis
Binary function matching is a foundational problem in reverse engineering, critical for tasks such as identifying known libraries in embedded firmware, isolating vulnerable functions, and understanding code reuse. However, this seemingly straightforward task is complicated by the myriad of variations introduced during compilation, including different settings, optimization levels, and compiler versions. Existing function matchers often struggle to provide truly *relevant* matches, prioritizing structural similarity over the insights a reverse engineer actually needs. This talk introduces **RevDECODE**, a novel framework that addresses these limitations by shifting the focus from mere similarity to **relevance decoding** through the intelligent use of contextual information.
AI review
RevDECODE is legitimate academic research on a genuinely hard problem — the similarity-vs-relevance gap in binary function matching — with a principled solution (context-aware layered graph + Viterbi-style decoding) evaluated at scale against multiple baselines. The numbers are credible, the framing is honest about what it does and doesn't replace, and the Franken binary testbed shows methodological maturity. Not a flashy talk, but the kind of foundational tooling work that quietly makes every binary analyst's life better.