NDSS

Differential Training: A Generic Framework to Reduce Label Noises for Android Malware Detection

Jiayun Xu (Singapore Management University), Yingjiu Li (University of Oregon), Robert H. Deng (Singapore Management University)

A common problem in machine learning-based malware detection is that training data may contain noisy labels and it is challenging to make the training data noise-free at a large scale. To address this problem, we propose a generic framework to reduce the noise level of training data for the training of any machine learning-based Android malware detection. Our framework makes use of all intermediate states of two identical deep learning classification models during their training with a given noisy training dataset and generate a noise-detection feature vector for each input sample. Our framework then applies a set of outlier detection algorithms on all noise-detection feature vectors to reduce the noise level of the given training data before feeding it to any machine learning based Android malware detection approach. In our experiments with three different Android malware detection approaches, our framework can detect significant portions of wrong labels in different training datasets at different noise ratios, and improve the performance of Android malware detection approaches.