Pk. Campbell et al., EXPERIMENTS WITH SIMPLE NEURAL NETWORKS FOR REAL-TIME CONTROL, IEEE journal on selected areas in communications, 15(2), 1997, pp. 165-178
We demonstrate the practical ability of neural networks (NN's) trained
in supervised mode to extract useful control ''knowledge'' from a lar
ge, high dimensional empirical database, and then to deliver almost op
timal control in ''real time.'' In particular, this paper describes ex
periments with NN-based controllers for allocating bandwidth capacity
in a telecommunications network, This system was proposed in order to
overcome a ''real time'' response constraint, Two basic architectures,
each consisting of a combination of two methods, are evaluated: 1) a
feedforward network-heuristic combination and 2) a feedforward network
-recurrent network combination, These architectures are compared again
st a Linear programming (LP) optimizer as a benchmark, This LP optimiz
er was also used as a teacher to label the data samples for the feedfo
rward NN training algorithm, NN-based solutions are very accurate (sim
ilar to 98% of optimal throughput) and, in contrast to the algorithmic
approach, can be delivered in ''real time.'' It is found that while t
he ''human'' generated heuristics (greedy search optimization) fail to
find a solution in approximately 30% of cases, the best NN fails only
in 4.9% of cases, Moreover, it has been found that in spite of the ve
ry high dimensionality of the problem (55 inputs and 126 outputs), the
solution can be delivered by surprisingly compact NN's, with as littl
e as around 1000 synaptic weights. This proves that on this occasion t
he NN's were able to extract simple but powerful ''heuristics'' hidden
in the complex sets of numerical data.