The U.S. Army has applied machine learning to develop a new method for detecting cyber threats that evade a system's security mechanisms.
Army Research Laboratory (ARL) worked with Samsung and the University of Southern California, Riverside to develop the fully automated Context Learning-based Adversarial Protection system, the Army said Tuesday.
“Our machine learning based method is principled and based on the observation that prior evasion attacks exhibit themselves in sequences of packets that are uncommon in benign traffic," said Zhiyun Qian, a professor at the University of California, Riverside.
Qian added the method uses a context-based learning approach to gather data on a cyber attack's inherent features.
Kevin Chan, a researcher at ARL, said these stealthy cyber-attacks would manipulate packets to instigate a variety of behaviors that camouflage malicious activities. The research will run for two more years through ARL's Cybersecurity Collaborative Research Alliance.