A consistent theme
in our technology is the management of composite alternatives, and causal
reasoning. Fundamentally, an infrastructure is a composite entity. One can
reason from effects to causes to discover the state of an infrastructure, and
reason from causes to effects to estimate the effects of various scenarios
upon the infrastructure.
The
framework offered by our products enables us to easily develop applications
for:
 | Fault diagnosis: Observations of anomalous behavior in the
infrastructure evoke hypotheses as to its cause. Testing the predictions of
those causes, and matching them against the observations, reveals the most believable causes..
|
| Design: Possible designs can be generated automatically (adding
or improving nodes, links, etc.), and they can be tested against various
situations (e.g., a link failing) to determine their merit. The best designs
are presented to the user, so they can examine the tradeoffs between
the infrastructure's costs and its robustness against various contingencies.
|
| Recovery and hardening: In response to a feared or an actual
attack, an infrastructure can be designed starting from its current state,
instead of from scratch, with the costs being calculated as an increment from the
existing equipment. So, as well as assisting in the design of new
infrastructures, we can bring value to the improvement of existing
infrastructures.
|
| Network-of-networks: Different infrastructures interact with one
another. For example, the power network can be instrumental to the operation
of the transportation network and to the distribution of water and fuel. Our
modeling language supports the ability to encode different interacting
infrastructures, so that our system can take account of a long chain of
effects as each problem with an infrastucture affects the others.
|
| Vulnerability analysis: Possible attacks against or failures of
parts of an infrastructure can be generated and their predictions tested.
Those of greatest predicted severity can then be studied by the user. This
is relevant both for planning attacks against others, and preparing for
attacks against oneself: for instance, denying communications to the enemy
while assuring one's own communications.
|
Based on previous Air Force
work, and with funding from the Army and the Navy, we examined the
vulnerability of a communications network to attacks on the nodes and the
links. An important question was of the command centers remaining in touch
with one another. Direct attacks on the command centers were costly, so
multiple targets around them had to be destroyed in order to isolate them. We
applied an evolutionary algorithm that progressively developed increasingly
complex attacks and successfully discovered a number of the attack plans that
caused the most impairment to communications for the least cost.
