Autonomic
Computing is an emerging area of study into the design and construction
of self-managed computing systems with a minimum of human interference[1].
Increasingly, DoD systems must operate with assured performance in
dynamic environments and under conditions not known a priori. Sources
of uncertainty include: multi-mode or simultaneous failures; ad hoc
composition of systems and networks; the inherently dynamic and uncertain
nature of malicious faults; the use of adaptive software; changing
system requirements; emergent behavior; and the growing recognition
that we can not specify, in advance, all system operating conditions
and functions.
The Advanced Computing
group at ALPHATECH is addressing autonomic computing through the research
and development of autonomic control systems. Autonomic control is
a specific approach to autonomic computing that employs feedback control
to enable automated tradeoffs between the failure cost of a compromised
information system and the maintenance cost of ongoing reactive countermeasures.
Scalable systems employing autonomic control will be able to adapt
their behavior, function, security posture, and performance in response
to uncertain dynamics while maintaining a desired or required measure
of performance. In a word, they will be self-managing. This approach
is multi-disciplinary - requiring advances in high assurance architectures,
machine learning, optimization, agent-based computing, and stochastic
control. Further, it requires tackling, simultaneously, the challenges
of trustworthiness, real-time guarantees, availability, reliability,
security, quality of service, and survivability.
[1] P. Horn, "Autonomic
Computing", IBM Research, http://www.research.ibm.com/autonomic/.
Autonomic Defense:
Thwarting Automated Attacks through Real-time Feedback Control
aLADS
takes a feedback control approach to address system survivability,
emphasizing the real-time sensing of locally apparent intrusive activity
with automated subsecond response selection. A prototype autonomic
defense system protects a Linux-based server from known and previously
unknown Internet worm attacks through the automated, real-time orchestration
of several operating system-based anomaly detectors and a variety
of system controls.
Autonomous,
Adaptive, and Cooperative Agent-based Systems for Unmanned Operations:
ALPHATECH
develops and maintains an Open Experimentation Framework for the Taskable
Agent Software Kit program which: supports investigation of solutions
in dynamic control and adaptation of large-scale multi-agent system
(MAS) behavior, provides a means to analyze agent/MAS designs with
respect to their ability to predict and control emergent behavior;
and provide a rich context in which to demonstrate applicability and
utility of multi-disciplinary MAS research to solve challenging dynamic
command and control problems.
Dynamic Control
and Formal Models of Multi-Agent Interactions & Behaviors:
ALPHATECH
research includes developing reinforcement learning (RL)-based agents
within a coherent mathematical framework that enables rigorous analytical
and empirical evaluation of a RL-based multi-agent system. Our focus
is on control, coordination, and adaptation of MAS behavior in uncertain
environments.
