Application oriented automatic structuring of time-delay neural networks for high performance character and speech recognition

Highly structured artificial neural networks can be optimized in many ways, and must be optimized for optimal performance. A highly structured approach is the multistate time delay neural network (MSTDNN) which uses shifted input windows and allows the recognition of sequences of ordered events that have to be observed jointly. An automatic structure optimization (ASO) algorithm is proposed and applied to MSTDNN-type networks. The ASO algorithm optimizes all relevant parameters of MSTDNNs automatically and is successfully tested with three different tasks and varying amounts of training data.<
>