[Abstract]As a feedforward neural network, the width learning system has attracted much attention from researchers because of its high accuracy, fast training speed and its ability to effectively replace deep learning methods. . However, the width learning system has problems such as being sensitive to the number of feature nodes in the network, and the pseudo-inverse method can easily cause the model to overfit. To this end, Bayesian inference and graph regularization are introduced in the width learning system. On the one hand, Bayesian learning by introducing prior knowledge can effectively improve the sparsity of weights and improve the stability of the model; on the other hand, adding graph regularization can fully consider the inherent graph information of the data and further improve the generalization of the model. ability. The performance of the proposed model is evaluated on the UCI data set and the NORB data set. The experimental results show that the proposed Bayesian width learning system model based on graph regularization can further improve the classification accuracy of the width learning system and has better stability.
[Keywords]Width learning system; Bayesian inference; graph regularization; pattern recognition
0 Preface
At present, the use of artificial intelligence technology to quickly and accurately obtain data for analysis and processing has become one of the issues of widespread concern. Neural network (NN) can extract effective features from massive data. Common models include radial basis function (RBF) network, self-organizing map (SOM) network, and recurrent neural network. Network (recurrent neural network, RNN), Boltzmann machine (BM), deep belief network (deep belief network, DBN), convolutional neural network (CNN), etc. However, in order to effectively complete more complex tasks, there are a large number of hyperparameters that need to be adjusted in the neural network model, which will lead to a longer training time and an overly complex structure of the model.