A Deep Dictionary Learning Model Based On Tensor

With the popularity of the Internet and its continued growth of scale,The form of characterization of data is becoming more and more complicated.Most of the traditional algorithms are expanding the data into one-dimensional vector,and then use the vector-based algorithm for data processing.However,that does not only undermine the data structure,but also brings some trouble for the subsequent in-deep study.In order to overcome the drawbacks of vectorized data,tensor-supported algorithms are increasingly replacing vector-based algorithms.Here,a new deep dictionary learning algorithm based on tensor is proposed to deal with tensor data directly and extract feature information.In this paper,we first introduce the characteristics of the linear combination and the translational invariance in the basic definition of tensor.According to these we propose a tensor-based dictionary learning(T-DL)algorithm and gives its specific calculation steps.What's more,we do the noisy comparison experiment on the Columbia University MSI data set between this new algorithm and TenSR algorithm.It is verified that the algorithm achieves the same or even better results,and the running speed is compressed a lot.Then,a new tensor-based Greedy Deep Dictionary Learning algorithm(T-GDDL)is proposed,and the concrete calculation step is given.Combined with the KNN classification algorithm,we compared it with the T-DL algorithm and other deep learning DBN and SAE algorithm to do the classification experiment.And it is verified that the T-GDDL algorithm is more accurate for the classification of MNIST handwritten data,and the operation time is faster than other deep models.Finally,A network fault detection system based on tensor is designed and implemented for the multimodality of the time,space and attribute of the current network signal.The tensor operation is used to reduce the dimension data of the tensor network,and the fault detection is effectively reduced.Time space complexity,but also improve the correctness of fault detection,is conducive to maintaining network security.We make a certain degree of expansion to the public data set KDD CUP99: add random time characteristics to constitutes a third-order network data model with time axes,sample points and sample attributes.The T-GDDL algorithm proposed in this paper is used to extract the characteristics of the network data.And then the SVM algorithm is used to do network fault detection by calculating the data after the feature extraction.Compared with the tensor and the dimensionality of the vector after dimensionality reduction,it is proved that the new method has better fault detection effect.Experiments show that the tensor-based deep dictionary learning algorithm in this paper is based on the tensor operation,so the information loss in the operation process is less.Therefore compared with other vector-based deep model algorithm,it has a higher accuracy.