Classifying Sequences of Extreme Length with Constant Memory Applied to Malware Detection

May 2021 Edward Raff, William Fleshman, Richard Zak, Hyrum S. Anderson, Bobby Filar, Mark McLean AAAI 2021

Most sequence models scale memory with sequence length — a problem when the sequence is a full PE binary. We present a constant-memory architecture that lets us train classifiers directly on whole binaries without truncation, and apply it to static malware detection.

AAAI 2021 · OJS