Homotopic BP algorithm adopted staged principle

The BP algorithm has solved the learning problem of multilayer feedforward networks successfully. But a lot of its applications indicated that it has the drawbacks of narrow convergence area, large learning cost, etc. The main motivation of neural network study is to learn from or simulate the architecture and the information processing mechanism of the human brain. Thus except for the adopted qualities of parallel processing, distributed storage and plasticity function, the efficient learning mechanism of the human brain should be considered in the ANN, too. On the other hand, how to use the priori knowledge or expert´s knowledge during network learning to achieve the transition form simplicity to complexity, hasn´t been solved well Since this algorithm adopts the topological homotopic theory, a feasible theoretical analysis tool for the feedforward network can be provided. Primary experiments show its good performance.