The Rise of Deep Learning for Detection and Classification of Malware

This McAfee research blog reports applying convolutional neural networks directly to raw PE bytes for malware detection and multi-family classification, describing experiments on large datasets (833,000 samples, deduped to 262,000; a 130,000-sample 11-class test) with high ROC/accuracy (up to 0.9953 AUC and 0.97 multi-class accuracy), plus t-SNE/PCA visualizations, XAI heatmaps highlighting influential byte sequences, and human analyst validation showing early identification of some families such as Emotet and Sodinokibi.