Neural Network Model Based On Clustering

The training of traditional neural network models requires a large amount of tagged data.In the era of big data,it takes time and effort to obtain tagged data.Considering that clustering algorithms can get the tags of data,can clustering and neural networks be combined to design a new model? This thesis will explore the content.By analyzing the objective function of clustering and the training target of neural network,we find that the optimization objectives are the same,and we can combine them together to construct a neural network model based on clustering.We first integrate unsupervised learning into neural networks to complete the construction of unsupervised neural network model based on clustering,and give the specific algorithm of this framework.Then,use clustering integration to make confidence on the clustering results.In the process of screening,the consensus design function proposed in this thesis is used to optimize the results of unsupervised learning of this framework.In order to further improve this framework,we add semi-supervised information to the model.The addition of semi-supervised information is mainly achieved in two ways.The one is to use semi-supervised clustering algorithms to replace the ordinary clustering algorithm into the framework,and the another is to use a semi-supervised clustering design consensus function in the cluster integration phase.At the end of the thesis,we use the standard dataset to compare the proposed algorithm framework with related algorithms.The experimental results show that compared with the general clustering algorithm,the unsupervised neural network model based on clustering has great advantages.The semi-supervised neural network model based on clustering obtained by adding semi-supervised information to the model can get better results.It can be seen that the model proposed in this thesis is better.