This patent deals with a hybrid local, global
mechanism for admission control in a communications network but is more
generally applicable to any network system where local control decisions
are made upon local sensory information but a global view would allow for
enhanced decision making. The patent proposes the use of search algorithms
from (population based) Evolutionary Computation for global search of all
allowed admission control strategies, coupled with local adaptation of
admission control strategies using Reinforcement Learning (or hybrids).
Local controllers perform online learning or adaptation. The global controller
performs off-line learning. Strategies are exchanged between local and
global controllers, allowing the global controller to compute the utility
of a particular strategy in a network-wide setting.
Please note: this patent has also been filed in
Canada, Europe and Japan.