Colleges Input-output Performance Appraisal Research Based On Feature Extraction Of Deep Learning

Deep learning is an important breakthrough of machine learning research,aims at building a deep network structure which can simulate the human brain to analysis and learn,It simulate the graded abstract feature representation strategy to interpret data,and has outstanding feature learning ability.Deep learning framework,which was proposed by Hinton et al.,is a multi-layer neural network built on unsupervised data;it uses greedy graded unsupervised way to resolve parameters optimization problem in deep neural network.Stacked Restricted Boltzmann Machines(RBMs)is one component module of the deep learning framework,it can effectively extract the data characteristics,and avails subsequent processing tasks such as classification,prediction and ratiocination.Thus,the RBMs is utilized as the construction module in proposed method.In the social background of comprehensively exploit market economy effect and promote performance management,college performance evaluation as a college management tool,it represents the time's request that higher education transfer from quantitative to qualitative leap development.An efficient performance evaluation can guide colleges reasonably distribute and adjust input resources and output effect.On the other hand,the comparison relationship between input feature and output feature also can be used to evaluate college performance status.Thus,in allusion to input data with multi-feature modules,this paper use RBMs to structure modular deep learning model,mainly includes:(1)Summarize the development progresses,research status and superiority of deep learning,briefly explain several common-used deep learning model and provide the detail theory and training algorithm of RBMs.(2)As to the dataset with multi-feature modules,connect multiple RBMs and top-layer classifier to build multi-class model.The basic ideas are firstly use the RBMs to extract features,then use the top-layer classifier to class features,finally fine-tuning the whole network parameters.Describe the network structure in proposed model,forward propagation and small batch gradient descent algorithm method in detail.In addition,deduce the gradient update formula of supervised fine-tuning process.(3)Compared to the classical linear feature extraction method PCA,RBM is a new feature extraction.This paper compares the feature extraction ability of PCA and RBM by experiments,designs the initial parameters of the proposed model,and discusses the influence of the number of hidden layers and nodes to model effect.(4)Applying the proposed method to the college input-output performance evaluationdata,and taking a comparison experiment between the proposed method with the Softmax classifier,deep neural network and deep belief network.Results indicate that the proposed method can obtain better error convergence and feature representation ability and can achieve higher prediction accuracy,which demonstrate the feasibility and effectiveness of the proposed method in college performance evaluation.