Improving security through lightweight learned models

点击翻译按钮获取中文摘要

The insecurity of modern computer systems threaten the success of the Internet.  Malicious software, in the form of viruses, worms, and spyware are widespread, causing data loss, privacy compromises, and losses of service. Even more disturbing, phishing emails and web pages now automate social engineering attacks on users themselves.  These problems stem from the imperfections endemic in complex software and the limitations humans have when interacting with such systems.  If we are to meaningfully improve the security of computer systems, we need mechanisms and strategies that can succeed in the face of such limitations.My ongoing research program is centered around the idea that lightweight, learned models based upon observed behavior can be used to improve computer security.  Models of program behavior can be used to detect security violations; models of networks can be used to reduce the impact of malicious network use; models of users can detect the presence of unauthorized individuals.  In the future I hope to develop, evaluate, and prototype models that can provide flexible and robust protection against the ever changing threats facing computer users.