Divide-and-conquer learning and modular perceptron networks

摘要: A novel modular perceptron network (MPN) and divide-and-conquer learning (DCL) schemes for the design of neural networks are proposed. When a training process in multilayer falls into local minimum or stalls flat region, proposed DCL scheme is applied to divide current data region two easier be learned regions. The continues when self-growing its initial weight estimation constructed one newly partitioned Another will resume on original network. Data partitioning, estimating iteratively repeated until all completely by MPN. We evaluated compared MPN with several representative two-spirals problem real-world dataset. achieved better performance requiring much less presentations during phases, generalization performance, processing time retrieving phase.