Using byte transition probabilities to create a "Markov image" that highlights the statistical structure of the archive.
To extract deep features, the raw binary data of the .003 file (which is the third part of a split 7-Zip archive) must be transformed into a visual format: High_Shrilling_Brother.7z.003
The model compresses the massive amount of raw data into a high-dimensional vector (the "deep feature") that uniquely represents the file's content. Using byte transition probabilities to create a "Markov
The first layers of the network detect simple edges or textures; deeper layers detect complex patterns unique to specific file types or malware families. High_Shrilling_Brother.7z.003