Optimisation of feedforward neural networks

摘要

A kind of method and apparatus, the internal structure for optimizing feedforward neural network (6) are disclosed with concealing device. The internal state of dynamic declaration (2) carries out network during training (1). According to dynamic declaration (2), training can interrupt static interpreter clause (4) generation. This, which generates geometric primitive, indicates that obtaining for internal state network a part is very detailed. According to this display, internal structure can be changed, restore before training. Another aspect of the present invention is related to optimizing training with initializing different training process (29) online according to the feature of dynamic monitoring and refer to control condition.