| Project Description: |
The overall objective is to build detectors that are robust against
any instance of an attack class i.e., the attack-mimicking problem will be directly addressed. The implication being
that multiple detectors will be necessary to provide sufficient coverage. The immediate benefit of such a scheme is that the
system is highly modular i.e., a simple path for detector maintenance and upgrade. Irrespective of the specific core attack,
each detector has the same basic requirements: generality and temporal reasoning. That is to say, the required temporal horizon for
a class of attacks is identified by the learning algorithm. The classical approach to such a problem is through recurrent architectures.
In this work we are interested in the utilization of multi-agent models in which problem decomposition and therefore temporal sequence
learning is established through an economic framework.
|
