Collaborative Learning for Security and Repair in Application Communities
Abstract
We investigated techniques that enable a system to learn where it is vulnerable to an attack or programming error, then automatically generate and evaluate ways that it can thwart the attack or recover from the error to continue to execute successfully. The approach is designed to work for systems, such as existing standard information technology installations, that have large monocultures of identical applications. By sharing information about attacks, errors, and response and recovery strategies, the system can quickly learn which strategies work best. The end result is a system whose robustness and resilience automatically grow over time as it learns how to best adapt and respond to the attacks and errors that its components inevitably encounter.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 2011
- Accession Number
- ADA550360
Entities
People
- Jeff Perkins
- Martin Rinard
- Michael Ernst
Organizations
- Massachusetts Institute of Technology