UMBC Information Systems Department
Machine Learning for Malware:
Challenges and Progress
Dr. Edward Raff
Booz Allen Hamilton
Visiting Prof. UMBC Computer Science & Electrical Engineering
12:00-1:00 pm ET Wednesday, 17 February 2021
online via WebEx
Malware is an ever-growing problem, single malware families have caused billions in damages, and the first direct death attributed to malware taking down a hospital has occurred. To detect new malware, machine learning is a naturally attractive approach. However, malware poses a number of unique challenges that have slowed the progress of ML-based solutions. In this talk, we will look at the task of malware detection from byte-based analysis, why it poses many challenging machine learning research problems, and progress we have made on these tasks by taking some non-standard approaches to machine learning: building shallow and wide networks instead of deep, handicapping the features of our model to make it robust, and using literal compression algorithms (LZMA) to find similar content.
Edward Raff leads Booz Allen’s machine learning research group and supports clients in developing new ML solutions. His research includes cybersecurity, adversarial machine learning, fairness and ethics, fingerprint biometrics, and high-performance computing. In his spare time, he is the author of the JSAT machine learning library. He received his BS and MS in Computer Science from Purdue University and his Ph.D. in CS from UMBC. Dr. Raff is a Nvidia Deep Learning certified instructor, and Visiting Professor at UMBC.