Graph Convolutional Networks (GCNs) have achieved notable success in skeleton-based action recognition. However, research scholars face issues such as low discriminability between actions and inadequate model posture adaptability in the field of action recognition. Furthermore, existing methods tend to ignore the temporal dependencies between joint points, employing a static adjacency matrix across all frames, which restricts the spatiotemporal modeling capacity of GCNs. To address these limitations, we propose a Dual Adjacency Graph Convolutional Network (DA-GCN) training framework combined with reinforcement learning. Our framework dynamically adjusts the spatial positions of joint points in skeleton data using a reinforcement learning policy network. This approach directly improves the differentiation between different actions while minimizing the computational demands associated with feature extraction. Moreover, DA-GCN employs a dual adjacency matrix design, constructing channel-dependent and temporal-dependent matrices to capture the spatiotemporal adjacency dependencies of joint points effectively. To our knowledge, this is the first study to leverage reinforcement learning to directly increase distinctions within the original skeleton data. Experimental results indicate that our method achieves excellent performance in the Northwestern-UCLA, NTU-RGB+D, and NTU-RGB+D 120 datasets, validating its effectiveness in action recognition tasks.

In recent years, unsupervised linear regression has attracted attention for its ability to directly capture the mapping relationship between samples and targets. However, existing algorithms can only utilize limited information from a single view, which often leads to unsatisfactory results. To address this problem, we propose a regression clustering model based on multi-view information fusion, called Scalable Multi-view Regression Clustering. This model consists of two parts: intra-view information fusion and inter-view information fusion. In the first part, to capture the local correlations among samples, we propose constructing view-specific bipartite graphs. Unlike traditional single-view and multi-view clustering algorithms, we treat the weights of the bipartite graph as additional features of the samples, thereby directly incorporating the local manifold structure of the samples at the feature level. Furthermore, since the original features of the samples also contain valuable information, we perform unsupervised linear regression separately on the samples represented by the original features and those represented by the bipartite graph weights in each view. The results are then integrated in a weighted manner. In the second part, we propose adaptively weighting the clustering results from each view to capture complementary information across views, thereby enhancing clustering performance. This strategy not only avoids the bipartite graph alignment issue in multi-view clustering but also enables clustering with linear time complexity, making it effective for handling large-scale data. An iterative optimization algorithm is developed to update all variables alternately. Experiments conducted on benchmark datasets demonstrate the superiority of our proposed model.

Human action recognition (HAR) based on visible light videos is prone to expose users privacy. Infrared videos and radio signals are well known invisible sensing approaches for HAR with the constraint of privacy preservation. However, either of them holds some inherent weaknesses, which strongly hinder the HAR performance improvement. Considering the complementarity of the different modal data, we suggest to explore both infrared and radio clues to improve the performance of HAR. Specifically, we design a two-branch heterogenous deep neural network to extract IR and WiFi features respectively. Rather than traditional methods, we propose a deep feature fusion approach which nonlinearly projects IR and WiFi features into a common subspace to maximize the correlation between two modalities. To validate our novel ideas, we build a human action dataset comprising IR and WiFi samples. Based on this dataset, a series of elaborate experiments were carried out, which prove that multi modal sensing can significantly improve HAR performance with the constraint of privacy preservation. The proposed feature fusion strategy outperforms other state-of-the-art methods obviously in experiments.

How to accurately locate the information required by researchers in the massive patent resources is an urgent problem to be solved in the patent database. Traditional collaborative filtering algorithms, mainly based on the users' rating scores, ignore their interests or direction changes and the active user's effect on recommended precision. To solve these problems, this paper reconstructs the recommendation weight with the consideration of time factor and item similarity, and proposed a practical method for calculating the similarity of item. This method can obtain a more reliable item similarity and better-recommended results by weakening the influence of active users on similarity calculation. The experimental results show that the proposed method obviously improves the recommendation performance. Compared with several traditional recommendation algorithms and deep learning algorithms in five datasets, the precision and recall rate improve by 5.2% and 9.5% on average, respectively.

The rapid advance in science and technology is made possible by research conduct and breakthroughs in a wide range of fields, which have resulted in a large number of academic papers. Searching through the enormous literature to find relevant information of one's research interest has become an increasingly important yet challenging problem for many researchers. Most existing methods for academic paper recommendation are based on the analysis of paper contents and only meet with limited success. We propose a novel method based on heterogeneous information networks for academic paper recommendation, referred to as HNPR. This method considers the citation relationship between papers, the collaboration relationship between authors, and the research area information of papers to construct two types of heterogeneous information networks. In such networks, a random walk-based strategy is used to simulate natural sentences for the discovery of relevance between two papers according to a mature natural language processing model. Extensive experimental results using real data in public digital libraries show that HNPR significantly improves the accuracy of academic paper recommendation in comparison with traditional content-based recommendation methods.

The past decade witnesses a great progress of personalized recommendation systems, which extract potential characteristics of users and items based on users' ratings information on items, and then recommend other items that may be of interest to users. However, traditional recommendation algorithms ignore the users social relationships and suffer with the cold start problem. In this paper, we propose a recommendation algorithm based on similarity calculation matrix factorization (SCMF) with the consideration of users social relationships and items associated relationships when ratings information is available. At the same time, when ratings information is unavailable, the proposed algorithm can obtain potential feature relationships and makes personalized recommendations. We also introduce a penalty factor of positive items to the similarity calculation equation in case of over fitting. By this way, the SCFM algorithm can improve the prediction accuracy and alleviate the problem of cold start of sparse data. We compared our approach with other 4 algorithms and conducted experiments on 4 common used data sets. Regarding the experimental results, the recommendation precision of SCMF can be improved by 2% to 9% compared with the traditional best algorithm, and by 0.17% to 18% for cold start users. Additionally, our approaches were applied to patent data resource exploitation, provided by Wanfang patent retrieval system. The experimental results also verify the effectiveness of SCMF algorithm in improving recommendation performance.

via principal component analysis (PCA), the LDA algorithm can avoid the small sample size problem. Most existing supervised dimensionality reduction methods extract the principal component of data first, and then conduct LDA on it. However, "most variance" is very often the most important, but not always in PCA. Thus, this two-step strategy may not be able to obtain the most discriminant information for classification tasks. Different from traditional approaches which conduct PCA and LDA in sequence, we propose a novel method referred to as joint principal component and discriminant analysis (JPCDA) for dimensionality reduction. Using this method, we are able to not only avoid the small sample size problem but also extract discriminant information for classification tasks. An iterative optimization algorithm is proposed to solve the method. To validate the efficacy of the proposed method, we perform extensive experiments on several benchmark data sets in comparison with some state-of-the-art dimensionality reduction methods. A large number of experimental results illustrate that the proposed method has quite promising classification performance.

Multi-view clustering has played a vital role in real-world applications. It aims to cluster the data points into different groups by exploring complementary information of multi-view. A major challenge of this problem is how to learn the explicit cluster structure with multiple views when there is considerable noise. To solve this challenging problem, we propose a novel Robust Auto-weighted Multi-view Clustering (RAMC), which aims to learn an optimal graph with exactly k connected components, where k is the number of clusters. ℓ1-norm is employed for robustness of the proposed algorithm. We have validated this in the later experiment. The new graph learned by the proposed model approximates the original graphs of each individual view but maintains an explicit cluster structure. With this optimal graph, we can immediately achieve the clustering results without any further post-processing. We conduct extensive experiments to confirm the superiority and robustness of the proposed algorithm.

In recent years, unsupervised two-dimensional (2D) dimensionality reduction methods for unlabeled large-scale data have made progress. However, performance of these degrades when the learning of similarity matrix is at the beginning of the dimensionality reduction process. A similarity matrix is used to reveal the underlying geometry structure of data in unsupervised dimensionality reduction methods. Because of noise data, it is difficult to learn the optimal similarity matrix. In this letter, we propose a new dimensionality reduction model for 2D image matrices: unsupervised 2D dimensionality reduction with adaptive structure learning (DRASL). Instead of using a predetermined similarity matrix to characterize the underlying geometry structure of the original 2D image space, our proposed approach involves the learning of a similarity matrix in the procedure of dimensionality reduction. To realize a desirable neighbors assignment after dimensionality reduction, we add a constraint to our model such that there are exact [Formula: see text] connected components in the final subspace. To accomplish these goals, we propose a unified objective function to integrate dimensionality reduction, the learning of the similarity matrix, and the adaptive learning of neighbors assignment into it. An iterative optimization algorithm is proposed to solve the objective function. We compare the proposed method with several 2D unsupervised dimensionality methods. K-means is used to evaluate the clustering performance. We conduct extensive experiments on Coil20, AT&T, FERET, USPS, and Yale data sets to verify the effectiveness of our proposed method.