The Holy Grail of "Systems for Machine Learning": Teaming humans and machine learning for detecting cyber threats

摘要: Although there is a large corpus of research focused on using machine learning to detect cyber threats, the solutions presented are rarely actually adopted in real world. In this paper, we discuss challenges that currently limit adoption security operations, with special focus label acquisition, model deployment, and integration findings into existing investigation workflows. Moreover, posit conventional approach development models, whereby researchers work offline representative datasets develop accurate not valid for many cybersecurity use cases. Instead, different needed: integrate creation maintenance models operations themselves.