Research on Transfer Learning of Multi-layer Neural Network against Network Data.

In the confrontation training of generative confrontation network, insufficient target sample training set will cause the model to fail to accurately learn the corresponding features, but it is difficult to obtain the target sample training set that needs to be manually made and marked. In response to this problem, the paper proposes a two-layer generative confrontation network model based on multi-layer neural network migration learning. In the first layer of the network, the model learns the approximate distribution of the target sample in the structural space through the pseudo-target sample. The learning ideas are transferred to the model and adjusted according to a small number of target samples in the second layer network. In the experiment, the paper verified the improvement of the model in Chinese font generation and picture frame conversion, and effectively trained a better model in a small number of target sample training sets.