CNN-AutoMIC: Combining convolutional neural
network and autoencoder to learn non-linear features for knn-based
malware image classification

The research report “CNN-AutoMIC: Combining convolutional neural network and autoencoder to learn non-linear features for knn-based malware image classification” has
been published on the prestigious first-tier magazine Computer Networks
.

The research, carried out by colleagues Andrea Iannacone, Simone Andriani and Antonio Maci of the Cyber Lab of Grottaglie – Rutigliano, in collaboration with the University of Bari, was born as part of the activities envisaged by the “Cybersecurity and SOC Products Suite” Program Contract.

The document presents an innovative algorithm for 0-day malware detection based on the classification of images generated from the binary representation of a file, using the combination of different ML techniques. A convolutional neural network (CNN) extracts features from the image, an autoencoder synthesizes them into a two-dimensional representation and a KNN classifier verifies whether or not they belong to the malware category, all following a previous training phase.

CNN-AutoMIC demonstrates superior classification performance compared to other state-of-the-art algorithms, without having high complexity and presenting absolutely reasonable training and classification times. The algorithm has already been integrated into the components of the BV TECH CyberSuite.