In several modular neural network (MNN) architectures, the individual decis
ions at the module level have to be integrated together using a voting sche
me. All these voting schemes use the outputs of the individual modules to p
roduce a global output without inferring explicit information from the prob
lem feature space. This makes the choice of the aggregation procedure very
subjective. In this work, a new MNN architecture will be presented. This ar
chitecture integrates learning into the voting scheme. We will be focusing
on making the decision fusion a more dynamic process. In this context, dyna
mic means the aggregation procedure which has the flexibility to adapt to c
hanges in the input. This approach requires the aggregation procedure to ga
ther information about the input to help better understand how to dynamical
ly aggregate decisions. (C) 1999 Published by Elsevier Science B.V. All rig
hts reserved.