Scalable Real-Time Deep Neural Network Optimization for Malware Classification

This project will investigate the possibility of training deep neural networks to detect malware using machine activity data during the execution of windows executable files. The novel element of the study will be to investigate the possibility of capturing the overall neuron activation state of the entire deep neural net and to identify areas of the net that can be optimized through recurrent feedback across the entire deep neural net using innovative methods beyond back propagation. Furthermore, this activity will be applied at scale on real-time sample throughput, so the project will also investigate methods for innovative input data reduction techniques and parallel training methods such as those supported by Apache SPARK and Microsoft's Cognitive Toolkit. The outcomes of the project will lead to novel computer science articles.