NEC Labs - Areas of Research

Robust and Secure Systems

Research Projects
Publications
Staff

Robust and Secure Systems

The ultimate survivable system (consisting of computing, storage, and networking resources) autonomously carries out its mission (described in terms of high-level objectives) and protects itself against known as well as unknown faults and threats. In case of a potential failure, the system has the ability to heal itself to fulfill its mission. The system constantly monitors changes in the environment, learns from experience, and adapts itself to become ever more robust.

Survivability has traditionally been provided mainly in mission- or safety-critical environments. As computing and networking become more pervasive, survivability becomes more crucial. Systems grow at a high rate increasing in both size and complexity but need to be managed by a constant number of administrators. Therefore, a pervasive infrastructure must be made trouble-free — a major research challenge. Unlike a closed system where redundancy has been engineered, a robust system cannot always be carefully planned. Instead, it evolves over time as components are removed or added.

Current efforts to build robust systems typically involve replicating critical resources and programming in advance mechanisms for detecting potential errors or attacks, and associated recovery procedures. The resilience of the system depends on its designers' ability to anticipate all likely future scenarios and on the scope and accuracy of the responses they programmed. This imperative computing approach to survivability is useful but has limitations in that it does not cope well in environments with frequent changes. Systems need to be able to adapt and evolve through learning and reasoning on their own.

Augmenting a system with cognitive capabilities for survivability is a promising prospect; however, many significant challenges lie ahead. A key issue is to come up with problem formulations and knowledge representations suitable for the system survivability domain. Examples of tasks in a survivable system: anomaly detection and fault diagnosis, responding (prevention, detection, reaction, and recovery) to intruders and worms, self-configuration and optimization. Another major challenge is complexity; it’s necessary to analyze the system to identify atomic problems that can be formulated as an automated learning problem. The real-time and mission-critical aspects of the robust systems domain also require several major shifts in machine learning focus: from supervised to unsupervised learning, from off-line to on-line learning, from fixed to dynamic environment, from centralized to distributed learning. Learning speed is a critical metric; unless learning techniques can be developed that can meet the accuracy/speed threshold, the technology is not practical in real-world systems.

NEC Labs America pursues the following research themes in Robust and Secure Systems: survivable grid storage, cognitive capabilities in survivable systems and fault detection in complex systems and applications.

Research Projects
Publications
Staff

NEC Laboratories America, Inc.
Princeton Campus - 4 Independence Way, Suite 200, Princeton NJ 08540
Cupertino Campus - 10080 North Wolfe Road, Suite SW3-350, Cupertino, CA 95014

webmaster@nec-labs.com