一句话：提出 MalCertain，利用 DNN 预测不确定性训练校正模型来修正检测结果，提升 Android 恶意软件检测准确率与对抗鲁棒性。
Android 恶意软件长期威胁推动了基于机器学习的检测研究，其中深度神经网络（DNN）方法因显著优于签名等传统方法而被广泛采用。然而，恶意软件快速演化且存在对抗样本，导致用早期样本训练的 DNN 模型在面对新出现的样本时决策质量较差。该现象本质上源于数据中的不确定性（噪声或随机性）与训练过程的不足（训练数据不充分），忽视这些不确定性会给模型预测带来风险。本文首次在恶意软件检测中估计 DNN 模型的预测不确定性，并利用该估计增强检测能力：除训练用于预测恶意软件的 DNN 外，还采用多种不确定性估计方法训练一个校正模型，判断 DNN 对某样本的预测正确与否，进而利用校正模型输出的不确定性来修正预测结果。实验表明，MalCertain 将底层 DNN 模型在 Android 恶意软件检测上的准确率提升约 21%，对对抗性 Android 恶意软件样本的检测效果提升最高达 94.38%。该工作揭示了利用预测不确定性改进基于预测的软件工程任务这一有前景的方向。
The long-lasting Android malware threat has attracted significant research efforts in malware detection. In particular, by modeling malware detection as a classification problem, machine learning based approaches, especially deep neural network (DNN) based approaches, are increasingly being used for Android malware detection and have achieved significant improvements over other detection approaches such as signature-based approaches. However, as Android malware evolve rapidly and the presence of adversarial samples, DNN models trained on early constructed samples often yield poor decisions when used to detect newly emerging samples. Fundamentally, this phenomenon can be summarized as the uncertainly in the data (noise or randomness) and the weakness in the training process (insufficient training data). Overlooking these uncertainties poses risks in the model predictions. In this paper, we take the first step to estimate the prediction uncertainty of DNN models in malware detection and leverage these estimates to enhance Android malware detection techniques. Specifically, besides training a DNN model to predict malware, we employ several uncertainty estimation methods to train a Correction Model that determines whether a sample is correctly or incorrectly predicted by the DNN model. We then leverage the estimated uncertainty output by the Correction Model to correct the prediction results, improving the accuracy of the DNN model. Experimental results show that our proposed MalCertain effectively improves the accuracy of the underlying DNN models for Android malware detection by around 21% and significantly improves the detection effectiveness of adversarial Android malware samples by up to 94.38%. Our research sheds light on the promising direction that leverages prediction uncertainty to improve prediction-based software engineering tasks.
@inproceedings{paperbot11354,
title = {MalCertain: Enhancing Deep Neural Network Based Android Malware Detection by Tackling Prediction Uncertainty},
author = {Haodong Li and Guosheng Xu and Liu Wang and Xusheng Xiao and Xiapu Luo and Guoai Xu and Haoyu Wang},
booktitle = {ICSE},
year = {2024},
doi = {10.1145/3597503.3639122}
}