Quick experiment to see if converting malware and non-malware binaries to B&W images and running them through a residual image recognition machine learning model has a good detection rate or not
Preliminary testing suggests it works pretty well, though I've not tested it with a vast range of malwares (trained with 30 or so specimens and tested with a handful of different ones).
You can try it out yourself here: https://ml-malware.onrender.com/ (takes a minute to load on first access)