Constructing Neural Networks by Extending the Optimization Field
Wenyuan Zhang · 2020
The performance of the neural network is not unilaterally increased with the training parameters of the neural network, and it is possible that the parameters of the training ground become more and the performance is degraded. This paper explains what the optimization field is and proposes a new way to solve this degradation problem. The new method is to built a neural network which includes the previous optimization field while adding parameters and extends the new optimization field. From this point, the paper explains the excellence of Inception, ResNet and DenseNet networks. Furthermore, this paper also proposes a new method for constructing neural networks that can improve the performance of neural networks with the increase of training parameters.