Aiming at the problem of low information utilization rate of sentence dependency tree and poor feature extraction effect in relation extraction task, an Attention-guided Gate perceptual Graph Convolutional Network (Att-Gate-GCN) model was proposed. Firstly, a soft pruning strategy based on the attention mechanism was used to assign weights to the edges in the dependency tree through the attention mechanism, thus mining the effective information in the dependency tree and filtering the useless information at the same time. Secondly, a gate perceptual Graph Convolutional Network (GCN) structure was constructed, thus increasing the feature perception ability through the gating mechanism to obtain more robust relationship features, and combining the local and non-local dependency features in the dependency tree to further extract key information. Finally, the key information was input into the classifier, then the relationship category label was got. Experimental results indicate that, compared with the original graph convolutional network relation extraction model, the proposed model has the F1 score increased by 2.2 percentage points and 3.8 percentage points on SemEval2010-Task8 dataset and KBP37 dataset respectively, which makes full use of effective information, and improves the relation extraction ability of the model.

Focusing on the issue that multi-view correlation analysis are not effective to exploit the correlation information and neglect latent discriminant information in images, a Dual-View Discriminant Correlation Analysis (DVDCA) approach based on dual view was proposed. Firstly, the supervised within-class correlation variation and between-class correlation variation were designed; secondly, within-class correlation variation was maximized and between-class correlation variation was minimized to extract the discriminant feature; finally, constrained dual-view discriminant correlation model was designed to exploit rich view information of both within-view and between-view. Compared with multi-view linear discriminant analysis, Canonical Correlation Analysis (CCA), Multi-view Discriminant Latent Space (MDLS), Uncorrelated Multi-view Discrimination Dictionary Learning (UMDDL) on the Multi-PIE dataset, the proposed algorithm can achieve recognition rate increase of 1.45-4.73 percentage points; on the MFD dataset, the proposed algorithm can achieve increase of 1.25-5.29 percentage points.

Concerning the problem that Fireworks Explosion Optimization (FEO) algorithm is easy to be premature and has low solution precision, an elite Opposition-Based Learning (OBL) was proposed. In every iteration, OBL was executed by the current best individual to generate an opposition search populations in its dynamic search boundaries, thus the search space of the algorithm was guided to approximate the optimum space. This mechanism is helpful to improve the balance and exploring ability of the FEO. For keeping the diversity of population, the sudden jump probability of the individual to the current best individual was calculated, and based on it, the roulette mechanism was adopted to choose the individual which entered into the child population. The experimental simulation on five classical benchmark functions show that, compared with the related algorithm, the improved algorithm has higher convergence rate and accuracy for numerical optimization, and it is suitable to solve the high dimensional optimization problem.

As the most commonly used method of reducing large-scale datasets, simple random sampling usually causes the loss of some clusters when dealing with unevenly distributed dataset. A density biased sampling algorithm based on grid can solve these defects, but both the efficiency and effect of sampling can be affected by the granularity of grid division. To overcome the shortcoming, a density biased sampling algorithm based on variable grid division was proposed. Every dimension of original dataset was divided according to the corresponding distribution, and the structure of the constructed grid was matched with the distribution of original dataset. The experimental results show that density biased sampling based on variable grid division can achieve higher quality of sample dataset and uses less execution time of sampling compared with the density biased sampling algorithm based on fixed grid division.

Chip layer assignment is a key step in analytical placement of 3D Integrated Circuits (ICs). Analytical placement could face the conversion from 3D continuous space in z-direction to several connected 2D chip layer spaces by layer assignment. However, layer assignment may destroy the previous optimal solution in 3D continuous space. To realize the transition from an optimal 3D placement to a legalized, layer-assigned placement smoothly, a layer assignment method was proposed by using the minimum cost flow, which protected solution space and inherited optimal wirelength at most. The layer assignment method was embedded in a multilevel non-linear placement of 3D ICs which minimized the weighted sum of total wirelength and Through Silicon Via (TSV) number subject to area density constraints. The proposed placement algorithm can achieve better wirelength results, TSV number and run time in comparison with the recent 3D placement methods.

To solve the path test data generation automatically in software testing, a new scheme on searching solution space based on Artificial Fish Swarm (AFS) algorithm was proposed. To improve the ability of original AFS, chaotic searching was introduced to reform AFS' local searching ability and precision of solution. Once AFS finished an iteration process, chaos algorithm was executed with global best solution. At the same time, some partial individuals with bad state were washed out. Then, according to the optimization individual contracting the searching space, some new individuals were generated randomly. Two kinds of triangle program were tested and the results show that the improved AFS has faster convergence and higher calculation accuracy.

To efficiently solve remote sensing image classification problem, a new classification algorithm based on manifold structure and Support Vector Machine (SVM) was proposed. Firstly, the proposed algorithm trained the SVM with initial training set and found the samples close to the decision hyperplane, then built the manifold structure of the samples by using Laplacian graph of the selected samples. The manifold structure was applied to find the representative samples for the classifier. The experimental evaluations were conducted on the hyperspectral images, and the effectiveness of the proposed algorithm was evaluated by comparing it with other active learning techniques exiting in the literature. The experimental results on data set confirm that the algorithm has higher classification accuracy.

Traditional GK clustering algorithm cannot automatically determine the number of clusters, and is sensitive to the initial cluster centers. According to these defects, an improved algorithm was proposed in this paper. Firstly, a new validity index, based on the weighted sum of separation between clusters and inter-cluster compactness, was proposed for the determination of the proper number of clusters. Then the idea of an improved entropy clustering was referenced to determine the initial cluster centers. Finally, the improved algorithm clustered the data sets according to the number of clusters given by the new index and the new cluster centers. The experimental results show that the new index works well in situations when there are overlapping clusters in the data set, and the improved algorithm has a higher clustering accuracy.

To correct the weakness such as sensitive to initial value of K, convergence untimely in K-Means algorithm, this paper presented an improved K-Means algorithm based on artificial fish school mechanism(NAFS). Firstly, priori knowledge is exploited to generate randomly some cluster centers for the problems to be solved, then composing the fish school environment. Secondly, the cooperation and competition mechanism of fish individuals is utilized to search satisfied outcome. In view of the deficiency that artificial fish school is prone to occur local optimum, niching algorithm is introduced according to the fish crowding density to ameliorate the diversity of population and improve its the solution accuracy. The results of experiments on KDDCUP99 show NAFS has higher clustering accuracy and is appropriate to solve clustering problems of high dimensionality.

These algorithms based on FP-tree, for mining maximal frequent pattern, have high performance but with many drawbacks. For example, they must recursively generate conditional FP-trees and many candidate maximum frequent itemsets. In order to overcome these drawbacks of the existing algorithms, an algorithm named Based on Dimensionality Reduction of Frequent Itemset (BDRFI) for mining maximal frequent patterns was put forward after the analysis of FPMax and DMFIA algorithms. The new algorithm was based on decreasing dimension of itemset. In order to enhance efficiency of superset checking, the algorithm used Digital Frequent Pattern Tree (DFP-tree) instead of FP-tree, and reduced the number of mining through prediction and pruning before mining. During the mining process, a strategy of decreasing dimension of frequent itemset was used to generate candidate frequent itemsets. The method not only reduced the number of candidate frequent itemsets but also can avoid creating conditional FP-tree separately and recursively. The experimental results show that the efficiency of BDRFI is 2-8 times as much as that of other similar algorithms.