Aiming at the problem that the existing Handwritten Mathematical Expression Recognition （HMER） methods reduce image resolution and lose feature information after multiple pooling operations in Convolutional Neural Network （CNN）， which leads to parsing errors， an encoder-decoder model for HMER based on attention mechanism was proposed. Firstly， Densely connected convolutional Network （DenseNet） was used as the encoder， so that the dense connections were used to enhance feature extraction， promote gradient propagation， and alleviate vanishing gradient. Secondly， Gated Recurrent Unit （GRU） was used as the decoder， and attention mechanism was introduced， so that， the attention was allocated to different regions of image to realize symbol recognition and structural analysis accurately. Finally， the handwritten mathematical expression images were encoded， and the encoding results were decoded into LaTeX sequences. Experimental results on Competition on Recognition of Online Handwritten Mathematical Expressions （CROHME） dataset show that the proposed model has the recognition rate improved to 40.39%. And within the allowable error range of three levels， the model has the recognition rate improved to 52.74%， 58.82% and 62.98%， respectively. Compared with the Bidirectional Long Short-Term Memory （BLSTM） network model， the proposed model increases the recognition rate by 3.17 percentage points. And within the allowable error range of three levels， the proposed model has the recognition rate increased by 8.52 percentage points， 11.56 percentage points， and 12.78 percentage points， respectively. It can be seen that the proposed model can accurately parse the handwritten mathematical expression images， generate LaTeX sequences， and improve the recognition rate.

Current main stream models cannot fully express the semantics of question and answer pairs， do not fully consider the relationships between the topic information of question and answer pairs， and the activation function has the problem of soft saturation， which affect the overall performance of the model. To solve these problems， an answer selection model based on pooling and feature combination enhanced BERT （Bi-directional Encoder Representations from Transformers） was proposed. Firstly， adversarial samples and pooling operation were introduced to represent the semantics of question and answer pairs based on the pre-training model BERT. Secondly， the relationships between topic information of question and answer pairs were strengthened by the feature combination of topic information. Finally， the activation function in the hidden layer was improved， and the splicing vector was used to complete the answer selection task through the hidden layer and classifier. Model validation was performed on datasets SemEval-2016CQA and SemEval-2017CQA. The results show that compared with tBERT model， the proposed model has the accuracy increased by 3.1 percentage points and 2.2 percentage points respectively， F1 score increased by 2.0 percentage points and 3.1 percentage points respectively. It can be seen that the comprehensive effect of the proposed model on the answer selection task is effectively improved， and both of the accuracy and F1 score of the model are better than those of the model for comparison.

In order to construct a large-scale containerized network， and achieve the purpose of building a high-fidelity， easy-to-program virtual network environment， a virtual network embedding algorithm based on time-varying resources was proposed to divide the OVS （Open vSwitch） and Docker based containerized network into segments and map them to several computing， network and storage resources constrained physical hosts. In the algorithm， firstly， the virtual network elements with close link relationships were aggregated hierarchically based on the topology of the virtual network to reduce the problem scale. Secondly， the importance scores of the aggregated virtual network nodes were obtained， the virtual network was segmented by the breadth first search algorithm and greedy strategy， and mapped into the physical hosts with suitable resources. Finally， the resource evaluation model in the algorithm was dynamically adjusted at runtime through the feedback at the fixed time of the resource consumption of the virtual network elements， so that the physical resources were effectively utilized. Experimental results show that the proposed algorithm can accommodate the virtual network with more than 1 300 network elements on multiple X86 hosts with low-level configuration， and can make the network jitter maintained at 0.1 ms or less.

Focusing on the prediction of student grade in the undergraduate teaching of higher education， a prediction algorithm based on course Knowledge Graph （KG） was proposed. Firstly， a course KG representing course information was constructed. Then， the neighbor-based methods and the KG representation learning-based methods were used to calculate the similarity of the courses on the knowledge level based on the KG， and those knowledge similarities among courses were integrated into the traditional grade prediction framework Collaborative Filtering （CF）. Finally， the performance of the algorithm with fusing KG and the common prediction algorithm in different data sparsities were compared in experiments. Experimental results show that in the data sparse scenario， compared with the traditional CF algorithm， the neighbor-based algorithm has the Root Mean Square Error （RMSE） reduced by about 11% and the Mean Absolute Error （MAE） reduced by about 9%； and compared with the traditional CF algorithm， KG representation learning-based algorithm has the RMSE reduced by about 17.55% and the MAE reduced by about 11.40%. Experimental results indicate that the CF algorithm using KG can significantly reduce the prediction error， which proves that the KG can be used as information supplement in the lack of historical data， thus helping CF to obtain better prediction results.

Concerning the problem that the network traffic data has been polluted by noise so that accurate modeling and predicting cannot be achieved, an integrated network traffic online predicting method based on lifting wavelet de-noising and online Least Squares Support Vector Machines (LSSVM) was proposed. First, the Lifting Wavelet De-noising (LWD) was used to pre-process network traffic data, then the phase space reconstruction theory was introduced to calculate the delay time and embedded dimension. On this basis, the training samples were formed and the online LSSVM prediction model was constructed to predict the network traffic. The experimental results show that this prediction model can eliminate the noise effectively and predict the network traffic.

Based on digital image scaling, a secure steganographic model for batch steganography was proposed, which conformed to the absolute safety definition. After being divided into many blocks, the secret information was deduced by using image scaling algorithm, rather than directly embedded in the carrier images. Firstly, a batch of carrier images were selected and zoomed to a specified magnification through a specific algorithm, then the relevance of the pixel information between secret image blocks and new images could be found out, finally new images were reduced to size of the original images for transmission. Since the scheme did not directly modify the image pixels, the original images were not required for extracting secret images and the security of the stegosystem had been improved. The experimental results and analysis show that the proposed algorithm is effective and it can be applied to image concealed communication.