Deep Learning for the Classification of Malicious Emails

Abstract

Deep learning has been widely used in many real world applications, including computer vision, object and image recognition, language translation and computer-aided medical diagnosis. Compared to conventional machine learning methods, deep learning has the advantage of being able to extract the features of raw data automatically, without using hand-tuned feature extractor. In this talk, we will first presenttwo methods of constructing our datasets of malicious emails, by considering the subject line andthe context of the malicious emails, and how to convert these datasets into suitable typeof training and test datasetsfor our numerical simulations. We then present the architecture of our convolutional neural network and classification accuracy for the datasets malicious emails.