A two-level learning hierarchy for constructing incremental projection generalizing neural networks

One of the incremental learning-based neural networks that theoretically guarantees the optimal generalization capability and provides exactly the same generalization capability as that obtained by batch learning is incremental projection generalizing neural networks. This paper will describe a two-level learning hierarchy for constructing the networks. An incremental projection learning in neural networks algorithm is employed at the lower level to construct the network while the learning parameters, the orders of the reproducing kernel Hilbert space, are optimized using a genetic algorithm at the upper level. The networks produced by this learning hierarchy will be used as subsystem of the artificial odor discrimination system to approximate percentage of alcohol.