The twin support vector machine improves the classification performance of the support vector machine by solving two small quadratic programming problems. However, this method has the following defects: (1) For the twin support vector machine and some of its variants, the constructed models use a hinge loss function, which is sensitive to noise and unstable in resampling. (2) The models need to be converted from the original space to the dual space, and their time complexity is high. To further enhance the performance of the twin support vector machine, the pinball loss function is introduced into the twin bounded support vector machine, and the problem of the pinball loss function not being differentiable at zero is solved by constructing a smooth approximation function. Based on this, a smooth twin bounded support vector machine model with pinball loss is obtained. The model is solved iteratively in the original space using the Newton-Armijo method. A smooth twin bounded support vector machine algorithm with pinball loss is proposed, and theoretically the convergence of the iterative algorithm is proven. In the experiments, the proposed algorithm is validated on the UCI datasets and the artificial datasets. Furthermore, the performance of the presented algorithm is compared with those of other representative algorithms, thereby demonstrating the effectiveness of the proposed algorithm.

Superpixels are an oversegmentation of an image and popularly used as a preprocessing in many computer vision applications. Many state-of-the-art superpixel segmentation algorithms rely either on minimizing special energy functions or on clustering pixels in the effective distance space. While in this paper, we introduce a novel algorithm to produce superpixels based on the edge map by utilizing a split-andmerge strategy. Firstly, we obtain the initial superpixels with uniform size and shape. Secondly, in the splitting stage, we find all possible splitting contours for each superpixel by overlapping the boundaries of this superpixel with the edge map, and then choose the best one to split it which ensure the superpixels produced by this splitting are dissimilarity in color and similarity in size. Thirdly, in the merging stage, the Bhattacharyya distance between two color histograms in the RGB space for each pair of adjacent superpixels is computed to evaluate their color similarity for merging superpixels. At last, we iterate the split-and-merge steps until no superpixels have changed. Experimental results on the Berkeley Segmentation Dataset (BSD) show that the proposed algorithm can achieve a good performance compared with the state-of-the-art superpixel segmentation algorithms.

Due to the complementary nature of graph neural networks and structured data in recommendations, recommendation systems using graph neural network techniques have become mainstream. However, there are still problems, such as sparse supervised signals and interaction noise, in the recommendation task. Therefore, this paper proposes a self-supervised graph attention collaborative filtering for recommendation (SGACF). The correlation between adjacent nodes is deeply mined using a multi-head graph attention network to obtain accurate node representations. It is worth noting that self-supervised learning is brought in as an auxiliary task in the recommendation, where the supervision task is the main task. It assists model training for supervised tasks. A multi-view of a node is generated by the graph data-augmentation method. We maximize the consistency between its different views compared to the views of the same node and minimize the consistency between its different views compared to the views of other nodes. In this paper, the effectiveness of the method is illustrated by abundant experiments on three public datasets. The results show its significant improvement in the accuracy of the long-tail item recommendation and the robustness of the model.

Density peaks clustering (DPC) algorithm can process data of any shape and is simple and intuitive. However, the distance between any two high-dimensional points tends to be consistent, which makes it difficult to distinguish the density peaks and easily produces “bad label” delivery. To surmount the above-mentioned defects, this paper put forward a novel density peaks clustering algorithm with isolation kernel and K-induction (IKDC). The IKDC uses an optimized isolation kernel instead of the traditional distance. The optimized isolation kernel solves the problem of converging the distance between the high-dimensional samples by increasing the similarity of two samples in a sparse domain and decreasing the similarity of two samples in a dense domain. In addition, the IKDC introduces three-way clustering, uses core domains to represent dense regions of clusters, and uses boundary domains to represent sparse regions of clusters, where points in the boundary domains may belong to one or more clusters. At the same time as determining the core domains, the improved KNN and average similarity are proposed to assign as many as possible to the core domains. The K-induction is proposed to assign the leftover points to the boundary domain of the optimal cluster. To confirm the practicability and validity of IKDC, we test on 10 synthetic and 8 real datasets. The comparison with other algorithms showed that the IKDC was superior to other algorithms in multiple clustering indicators.