IMPROVING MODEL ACCURACY USING OPTIMAL LINEAR-COMBINATIONS OF TRAINEDNEURAL NETWORKS

Neural network (NN) based modeling often requires trying multiple netw orks with different architectures and training parameters in order to achieve an acceptable model accuracy. Typically, only one of the train ed networks is selected as ''best'' and the rest are discarded. We pro pose using optimal linear combinations (OLC's) of the corresponding ou tputs of a set of NN's as an alternative to using a single network. Mo deling accuracy is measured by mean squared error (MSE) with respect t o the distribution of random inputs. Optimality is defined by minimizi ng the MSE, with the resultant combination referred to as MSE-OLC. We formulate the MSE-OLC problem for trained NN's and derive two closed-f orm expressions for the optimal combination-weights. An example that i llustrates significant improvement in model accuracy as a result of us ing MSE-OLC's of the trained networks is included.