Deep Neural Network-Based On-device Malware Detection (DOM)
DOM is a deep learning-based on-device malware detection application for Android devices that employs transfer learning and model personalization to improve a global deep neural network (DNN) across smartphones. DOM is lightweight, robust, and does not need to have smartphones rooted.
DOM analyzes applications on smartphones without a need for a remote server and trains a neural network on the device with the help of a generic DNN for labeling. DOM employs two on-device machine learning models called generic and personalized models and dynamically analyzes applications to extract a comprehensive set of features. The generic model is used to label applications whose ground truth is unavailable. In contrast, the personalized model is a lightweight trainable model created by retaining most of the generic DNN layers and trainable parameters and adding a new lightweight neural network, which is further improved with the help of federated learning.
DOM is an ongoing project and new features and improvements will be added. The current version is DOM-1.0.0.
If you use DOM or any parts of the DOM project, you must cite the following paper DOM's paper.
A. Pasdar, Y. C. Lee, T. Liu and S. -H. Hong, "Train Me to Fight: Machine-Learning Based On-Device Malware Detection for Mobile Devices," 2022 22nd IEEE International Symposium on Cluster, Cloud and Internet Computing (CCGrid), 2022, pp. 239-248, doi: 10.1109/CCGrid54584.2022.00033.