Malware classification beyond signature matching
MalVision
A research project that converts binaries into images and uses convolutional models to identify malware families beyond signature matching.
Current state
Research project and experimental classifier.
Case study
Problem
Signature matching does not generalize well to malware variants that have changed at the byte level.
Approach
The research pipeline represents binaries as images and learns visual patterns associated with malware families.
Engineering
Python handles the pipeline, with PyTorch and scikit-learn used for modeling and evaluation.