Deep Learning – Classifying malicious websites with image recognition models

Schedule Not Yet Finalized October 2, 2018 - Feedback     

Bookmark and Share

Akbar Qureshi

During this presentation I will demonstrate how convolutional neural network (CNN) models used for image recognition can also be used to classify malicious websites. I will go over how a CNN trained on images of botnet C2 panels and phishing websites can accurately predict and label, if a given image of a malicious website is a C2 panel or a phishing website. For this presentation, I have trained the CNN model to predict images of Pony and Lokibot botnet C2 panels as well as images of Facebook, Dropbox and Office365 phishing websites. I will also show a live demo of the proposed solution.