[2G1-ES-4] Robots and real worlds: Applied machine learning

[2G1-ES-4-01] On Identifying Functions in Binaries using Neural Networks

〇Seng Pei Liew¹, Takuma Amada¹ (1. NEC Corporation)

キーワード：Security, Binary analysis, AI

Identifying functions in binary codes is the very first step of performing binary analysis, and is known to be a difficult problem. In this work, we first show that neural networks are able perform this task with performance better than conventional methods. However, as is well known, deep learning models are vulnerable to adversarial examples. We also show in this work that, function identifying neural networks are vulnerable to adversarial inputs in the form of binary obfuscation.