Making & Breaking Machine Learning Anomaly Detectors in Real Life

No ratings

Presented at BSidesNYC 2016 by

Clarence Chio

Machine learning (ML) techniques used in network intrusion detection are susceptible to adversarial 'model poisoning'. We demonstrate this attack & analyze some proposals for how to circumvent it, then consider specific use cases of ML and anomaly detection in a security context. We analyze how ML can be used in security and why it is not as straightforward as once thought.