PERFUME

Abstract

Algorithmic identification is the crux for several binary analysis applications, including malware analysis, vulnerability discovery, and embedded firmware reverse engineering. However, data-driven and signature-based approaches often break down when encountering outlier realizations of a particular algorithm. Moreover, reverse engineering of domain-specific binaries often requires collaborative analysis between reverse engineers and domain experts. Communicating the behavior of an unidentified binary program to non-reverse engineers necessitates the recovery of algorithmic semantics in a human-digestible form. This paper presents PERFUME, a framework that extracts symbolic math expressions from low-level binary representations of an algorithm. PERFUME works by translating a symbolic output representation of a binary function to a high-level mathematical expression. In particular, we detail how source and target representations are generated for training a machine translation model. We integrate PERFUME as a plug-in for Ghidra--an open-source reverse engineering framework. We present our preliminary findings for domain-specific use cases and formalize open challenges in mathematical expression extraction from algorithmic implementations.