Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2020-09-28 Cooperative journals: 《计算机应用研究》
Abstract： This paper presented a novel discriminative K-SVD network (DKSVDN) for face recognition. It embedded discriminative information into traditional K-SVD algorithm by special design of dictionary as well as sparse representation coefficients on the dictionary. The dictionary consisted of label specific atoms and descriptive atoms, while sparse codes contained one-hot label vectors and descriptive codes. In addition, as sparse representation algorithms were time-consuming, DKSVDN attached a co-trained feed-forward neural network to discriminative dictionary learning model to predict sparse codes. Moreover, with generative module in DKSVDN, this work also designed a new dreaming training phase to improve the robustness of DKSVDN for unknown pattern in known class. The experiment results on public face image datasets verified effectiveness of this method.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-01-28 Cooperative journals: 《计算机应用研究》
Abstract： Sparse representation classification (SRC) has a good performance of classification when the feature space spanned by training samples is sufficient, but the computational cost is expensive. To solve this problem of SRC, this work considered the constraint of reconstruction samples. It introduced a group sparsity effect to enhance the competitions between different subjects in reconstruction procedure, and improves the accuracy of classification finally. Since the proposed method has a closed solution, the computational cost is very low. Moreover, the sparsity of the coefficient produced by the new approach is the same as that obtained by SRC. The experiment results on public face and object image datasets demonstrate that the proposed method has a good performance comparing with other same kind approaches.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-18 Cooperative journals: 《计算机应用研究》
Abstract： Multi-task and multi-kernel learning has gradually become the research focus of online learning algorithms. For the prediction of data stream, some online learning algorithms have some shortcomings in accuracy. Therefore, this paper proposes a new multi-task and multi-kernel online learning model to improve the accuracy of data stream prediction. Based on the multi-task multiple-kernel learning, we extends the model to online learning, so as to get a new online learning algorithm, while maintaining a certain size of the input data window for the integrity of the data with less space. In the experimental part, the selection of kernel function and the size of training sample set are analyzed in detail. Through the analysis of UCI data and actual airport passenger flow data, the algorithm proposed in this paper can ensure the accuracy and real-time of stream data processing, and has certain applicable value.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-02 Cooperative journals: 《计算机应用研究》
Abstract： Computer recognition of human emotion has become a hot topic in the field of brain computer interface (BCI) in recently years. By analyzing the various emotional states in people’s life, extracting the features of EEG and classifying emotional states is an important direction in the field of emotional intelligence. Based on the emotion data set induced by the music video, this research extracted the frequency-domain features of EEG. After that, the Accelerated Proximal Gradient (APG) and Orthogonal Matching Pursuit (OMP) algorithms for the sparse representation method were adopted to classify the EEG signals. By comparing with other algorithms, the experimental results show that the APG with L1 norm performs well in the emotion data set with fast convergence speed, and the greedy idea based OMP algorithm can achieve the same effect with other algorithms. The comparative analysis in this paper show the effectiveness and feasibility of the proposed method for emotional EEG signals classification.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-19 Cooperative journals: 《计算机应用研究》
Abstract： In the past, most of the traditional link prediction methods are oriented to the undirected network, in fact, most social networks are directional, and do not consider the duplication between the same node pair and the microscopic evolution information in the network, therefore they can not solve Link prediction in directed dynamic networks better. This paper focused on the directional network, the repeated edge information between the pair of nodes is transformed into the weight of the edge between the pair of nodes, then used the evolution model based on the triad motif, calculate the motif transformation probability matrix between the adjacent time slice in the move window, the probability matrix be analyzed by exponentially weighted moving average, and then it used the matrix to predict the chain edge in the network. This method not only makes full use of the network micro evolution information, but also solves the problem of overlapping edges in dynamic network. Experiments show that this method can get better results than Common Neighbor, Triad Transition Matrix and other methods in network with high Global Clustering Coefficient and high Average Degree．Therefore, this method can apply the network microscopic information to the link prediction better.
Hits
Hits
Downloads
Comment