SecML

From RAD Lab

Jump to: navigation, search

Contents

Learning in Security Sensitive Environments

People

Grads          Undergrads
  • Steven Lee
  • Chris Cai
             Alumni
Faculty              Collaborators

Summary: Security of Adaptive Systems

Machine learning is becoming prevalent in the systems domain as a detection and analysis tool for problems amenable to adaptive techniques. However, the adaptivity and flexibility that are machine learning's biggest assets are also qualities that an attacker might exploit. Thus, it is important to study the security of learning systems.

One research direction is to experimentally and theoretically analyze existing systems. In [2], the authors use PCA for detecting anomalous point-to-point flows based on link volume data. We are investigating the effect an adversary can have on the normal subspace of link volume vectors learned under various realistic models of control. In a similar vein, we are exploring the vulnerabilities of the spam filter, SpamBayes [3].

Another focus is security as a property of families of learners. Universal sequence prediction [1] considers the loss of a learner in the presence of an adversary. This approach is appropriate for security, as the adversary is modeled in a general way. Robust statistics is another appropriate framework, which quantifies the effect of outliers. For security, it is important to quantify the cost of an attack, possibly in the presence of non-adversarial data.

[1] N. Cesa-Bianchi and G. Lugosi. Prediction, Learning, and Games. Cambridge University Press, 2006.

[2] A. Lakhina, M. Crovella and C. Diot. Diagnosing network-wide traffic anomalies. ACM SIGCOMM Computer Communication Review, 34(4), 2004.

[3] T. A. Meyer and B. Whateley. SpamBayes: Effective open-source, Bayesian based, email classification system. Conference on Email and Anti-Spam, 2004.

Publications

Links