Security Lab Seminar
Sept. 4, 2002
1:30 pm
Yihua Liao
"Machine Learning Techniques for Mitigating Insider Threat"
Insiders constitute the greatest threat to an information system's
security. User profiling is an important technique for detecting an
insider's misuse. We present our preliminary results with the Windows NT user
profiling data.
Furthermore, we propose FMIT, a framework for mitigating the
insider threat. FMIT employs machine learning techniques to adaptively learn
user behavior profiles in real time. Meanwhile, user interests are inferred
from a user's Web activities, which in turn can help the
security officers understand a user's behavior and infer his or her intent.
FMIT overcomes the drawbacks of many previous user profiling
systems and provides a new observable for understanding user behavior.