Proj CDeepFuzz Paper Reading: PELICAN: Exploiting Backdoors of Naturally Trained Deep Learning Models In Binary Code Analysis

Abstract

背景：

1. 本文研究的不是被恶意植入的后门，而是products of defects in training
2. 攻击模式: injecting some small fixed input pattern(backdoor) to induce misclassification

本文: PELICAN
Github: https://github.com/ZhangZhuoSJTU/Pelican
Task: find backdoor vulnerabilities in transformer models of binary analysis
Method: 保留code snippet的原程序语义，同时让trigger成为语义的一部分从而不能轻易eliminate

实验：
datasets:

• Tasks: Disassembly, Function Signature Recovery, Function Name Prediction, Compiler Provenance, Binary Similarity
• models:Disassembly(BiRNN-func, XDA-func, XDA-call), Function Signature Recovery(StateFormer, EKLAVYA, EKLAVYA++), Function Name Prediction(in-nomine, in-nomine++), Compiler Provenance(S2V, S2V++), Binary Similarity(Trex, SAFE, SAFE++, S2V-B, S2V-B++)
• Commercial tools: DeepDi, BinaryAI

效果：

1. 能够在white-box and black-box 环境下成功诱发misclassification
2. 能在两个商业tools上找到backdoor vulnerabilities

本栏目推荐文章
• Early lameness detection in dairy cattle based on wearable gait analysis using semi-supervised LSTM-Autoencoder
• Understanding q-value and FDR in Differential Expression Analysis
• BigDataAIML-Kaggle-How to Calculate Principal Component Analysis (PCA) from Scratch in Python
• 【五期杨志】CCF-A（CVPR'22） Dual-Key Multimodal Backdoors for Visual Question Answering
• SR Algorithm Analysis（1）——ZSSR
• PCA(Principal Components Analysis)主成分分析: 一维列向量坐标的变换是左乘变换矩阵 而 一维行向量的坐标系基元的变换 是 右乘变换矩阵
• Overview of Machine Learning Methods for Genome-Wide Association Analysis
• Detecting Unknown Encrypted Malicious Traffic in Real Time via Flow Interaction Graph Analysis
• PCA (principal component analysis)算法
• Probabilistic principal component analysis-based anomaly detection for structures with missing data（概率主成分分析PPCA）

Exploiting CDeepFuzz Backdoors Naturally Analysisexploiting cdeepfuzz backdoors naturally naturally backdoors multimodal backdoors answering dual-key exploiting extraction exploiting practical response computation exploiting learning networks classification exploiting questions inference recommendation session-based information exploiting interactions exploiting high-order improving