The STAMINA whitepaper presents a scalable deep learning method for static malware detection using image-based transfer learning. Intel and Microsoft convert malware binaries into images, then apply pre-trained deep neural networks (like Inception-v1) to classify them. STAMINA achieves 99.07% accuracy with low false positives (2.58%) on small files and 95.97% accuracy for large files using segmentation. The model bypasses manual feature engineering and adapts to varied file sizes through file-size gating. It highlights strong potential for accurate, efficient malware classification, especially for static analysis scenarios.

Join for free to read