Malware Detection with Convolutional Neural Network Using Hardware Events

摘要: Detection of malicious programs (i.e., malwares) is a great challenge due to increasing amount and variety attacks. Recent works have shown that machine learning, especially neural network, performs well in malware detection. In this paper, convolution network (CNN) used build the classification model. Different from other works, our work uses hardware events generate feature image programs. These events, such as cache miss rate, branch misprediction can be collected performance counter Intel CPUs. We train CNN with kinds data sizes kernel sizes, evaluate result by area under receiver operating characteristics (ROC) curve (AUC). The results show proposed model achieve AUC = 0.9973 best case influence size or very little. Moreover, comparison CNNs trained software-based features, it indicated has higher accuracy than ones.