Search

Abstract: Traditional term weighting schemes in text categorization, such as TF-IDF, only exploit the statistical information of terms in documents. Instead, in this paper, we propose a novel term weighting scheme by exploiting the semantics of categories and indexing terms. Specifically, the semantics of categories are represented by senses of terms appearing in the category labels as well as the interpretation of them by WordNet. Also, the weight of a term is correlated to its semantic similarity with a category. Experimental results on three commonly used data sets show that the proposed approach outperforms TF-IDF in the cases that the amount of training data is small or the content of documents is focused on well-defined categories. In addition, the proposed approach compares favorably with two previous studies. [Copyright &y& Elsevier]

The efficiency of communication across workers is a significant factor that affects the performance of federated learning. Though periodic communication strategy is applied to reduce communication rounds in training, the communication cost is still high when the training data distributions are not independently and identically distributed (non-IID) which is common in federated learning. Recently, some works introduce variance reduction to eliminate the effect caused by non-IID data among workers. Nevertheless the provable optimal communication complexity O (log (S T)) and convergence rate O (1 / (S T)) cannot be achieved simultaneously, where S denotes the number of sampled workers in each round and T is the number of iterations. To deal with this dilemma, we propose an optimization algorithm SQUARFA that adopts stagewise training framework coupling with variance reduction and uses a quick-start phase in each loop. Theoretical results show that SQUARFA achieves both optimal convergence rate and communication complexity for both strongly convex objectives and non-convex objectives under PL condition, thus fills the gap mentioned above. Then, a variant of SQUARFA yields the optimal theoretical results for general non-convex objectives. We further extend the technique in SQUARFA to the large batch setting and achieve optimal communication complexity. Experimental results demonstrate the superiority of the proposed algorithms. [ABSTRACT FROM AUTHOR]

As one of the important psychological stress reactions, Micro-expressions (MEs) are spontaneous and subtle facial movements, which usually occur in a high-stake situation and can reveal genuine human feelings and cognition. ME, Recognition (MER) has essential applications in many fields such as lie detection, criminal investigation, and psychological healing. However, due to the challenges of learning discriminative ME features via fleeting facial subtle reactions as well as the shortage of available MEs data, this research topic is still far from well-studied. To this end, in this paper, we propose a deep prototypical learning framework, namely ME-PLAN, with a local attention mechanism for the MER problem. Specifically, ME-PLAN consists of two components, i.e., a 3D residual prototypical network and a local-wise attention module, where the former aims to learn the precise ME feature prototypes through expression-related knowledge transfer and episodic training, and the latter could facilitate the attention to the local facial movements. Furthermore, to alleviate the dilemma that most MER methods need to depend on manually annotated apex frames, we propose an apex frame spotting method with Unimodal Pattern Constrained (UPC) and further extract ME key-frames sequences based on the detected apex frames to train our proposed ME-PLAN in an end-to-end manner. Finally, through extensive experiments and interpretable analysis regarding the apex frame spotting and MER on composite-database, we demonstrate the superiority and effectiveness of the proposed methods. • We explore deep prototypical learning for the MER problem, and try to address three challenges in this topic. • We propose a novel deep prototypical learning framework, namely ME-PLAN, using a 3D residual prototypical network with episodic training and a local-wise attention module to learn precise ME representation. • A novel apex frame spotting method based on Unimodal Pattern Constraint is proposed to effectively eliminate noise interference and accurately locate apex frames. • Extensive experimental results with a comparison of state-of-the-art methods have demonstrated the effectiveness of our ME-PLAN on apex frame spotting and MER tasks. [ABSTRACT FROM AUTHOR]

With the prevalence of GPS-enabled smart phones, Location Based Social Network (LBSN) has emerged and become a hot research topic during the past few years. As one of the most important components in LBSN, Points-of-Interests (POIs) has been extensively studied by both academia and industry, yielding POI recommendations to enhance user experience in exploring the city. In conventional methods, rating vectors for both users and POIs are utilized for similarity calculation, which might yield inaccuracy due to the differences of user biases. In our opinion, the rating values themselves do not give exact preferences of users, however the numeric order of ratings given by a user within a certain period provides a hint of preference order of POIs by such user. Firstly, we propose an approach to model users preference by employing utility theory. Secondly, We devise a collection-wise learning method over partial orders through an effective stochastic gradient descent algorithm. We test our model on two real world datasets, i.e., Yelp and TripAdvisor, by comparing with some state-of-the-art approaches including PMF and several user preference modeling methods. In terms of MAP and Recall, we averagely achieve 15% improvement with regard to the baseline methods. The results show the significance of comparative choice in a certain time window and show its superiority to the existing methods. [ABSTRACT FROM AUTHOR]

Traffic flow prediction, a crucial application of intelligent transportation systems (ITS), has become an increasingly prevalent research topic. However, existing models that achieved high prediction accuracy on selected metrics may suffer from time delay in prediction curves which has been rarely explored. These models may produce seemingly accurate but invalid predictions by merely tracking and replicating previous true values. To address this anomaly, we propose a highly interpretable prediction mechanism, the Incremental Output Decomposition Recurrent Neural Network (IODRNN). We also introduce a new metric called Shift Divergence Difference (SDD) to assess the degree of latency of the overall sequence and evaluate the effectiveness of IODRNN in reducing the delay phenomenon. Our experimental results using real-world GNSS traffic data show that IODRNN has the smallest degree of latency and improves MAE and RMSE by 16.8% and 17.4% on average, respectively, over most contrast models. Our study presents an effective approach to evaluate prediction latency, ensuring validity and robustness in traffic prediction. [ABSTRACT FROM AUTHOR]

Highlights • A structure-enriched framework for network embedding from a holistic perspective. • Introducing the direction tuning parameters into multi-order transition matrices. • Using the denoise autoencoder to do the dimension reduction of multi-order matrices. • Employing the method with attention mechanism to combine multi-order information. Abstract Recent years have witnessed the importance of network embedding in many fields, as well as increased attention in academia. Although a number of algorithms have been proposed in this area, most existing models which only utilize the structure topology information of networks often suffer performance losses because of their insufficiency with regard to selecting structure similar patterns, handling noise data, and/or capturing non-linear or high-order structure information. To address these challenges, in this paper, we present a novel S tructure- E nriched N eural N etwork (SENN) for network embedding. Specifically, SENN can not only capture the complex structure similar patterns observed in networks by introducing direction adjustment parameters of the transition probability, but also introduce a stacked denoise autoencoder to perform the dimension reduction for each order matrix independently. Therefore, SENN can preserve more useful structure information and make the embeddings more robust. Moreover, SENN can effectively integrate the multi-order structure information by the combining layer with attention mechanism. Finally, to compare with other state-of-the-art methods, we conduct extensive experiments with both synthetic and real-world datasets on various tasks (e.g.,node classification, visualization). The experimental results clearly demonstrate the effectiveness of our proposed model for network embedding. [ABSTRACT FROM AUTHOR]

Most of existing text automatic summarization algorithms are targeted for multi-documents of relatively short length, thus difficult to be applied immediately to novel documents of structure freedom and long length. In this paper, aiming at novel documents, we propose a topic modeling based approach to extractive automatic summarization, so as to achieve a good balance among compression ratio, summarization quality and machine readability. First, based on topic modeling, we extract the candidate sentences associated with topic words from a preprocessed novel document. Second, with the goals of compression ratio and topic diversity, we design an importance evaluation function to select the most important sentences from the candidate sentences and thus generate an initial novel summary. Finally, we smooth the initial summary to overcome the semantic confusion caused by ambiguous or synonymous words, so as to improve the summary readability. We evaluate experimentally our proposed approach on a real novel dataset. The experiment results show that compared to those from other candidate algorithms, each automatic summary generated by our approach has not only a higher compression ratio, but also better summarization quality. [ABSTRACT FROM AUTHOR]

Online karaoke allows users to practice singing and distribute recordings. Different from traditional music recommendation, online karaoke need to consider users’ vocal competence besides their tastes. In this paper, we develop a karaoke recommender system by taking into account vocal competence. Alone this line, we propose a joint modeling method named MKLA by adopting bregman divergence as the regularizer in the formulation of multiple kernel learning. Specially, we first extract users’ vocal ratings from their singing recordings. Due to an ever-increasing number of recordings, the evaluations in large-scale kernel matrix may cost lots of time and internal storage. Therefore, we propose a sample compression method to eliminate users’ vocal ratings, exploit an MKL method, and learn the latent features of the vocal ratings. These latent features are simultaneously fed into a bregman divergence and then we use the trained classifier to predict the overall rating of a user with respect to a song. Enhanced by this new formulation, we develop the SMO method for optimizing the MKLA dual and present a theoretical analysis to show the lower bound of our method. With the estimated model, we compute the matching degree of users and songs in terms of pitch, volume and rhythm and recommend songs to users. Finally, we conduct extensive experiments with online karaoke data. The results demonstrate the effectiveness of our method. [ABSTRACT FROM AUTHOR]

• Modeling the market evolution for start-up projects via a graph-based model. • Modeling the competitiveness pressure by aggregating funding states of rivals. • Propagation tree structure with hierarchical updating can track the market evolution. • The proposed model achieved the best result compared with all powerful baselines. In the online innovation market, the fund-raising performance of the start-up project is a concerning issue for creators, investors and platforms. Unfortunately, existing studies always focus on modeling the fund-raising process after the publishment of a project but the predicting of a project attraction in the market before setting up is largely unexploited. Usually, this prediction is always with great challenges to making a comprehensive understanding of both the start-up project and market environment. To that end, in this paper, we present a focused study on this important problem from a market graph perspective. Specifically, we propose a Graph-based Market Environment (GME) model for predicting the fund-raising performance of the unpublished project by exploiting the market environment. In addition, we discriminatively model the project competitiveness and market preferences by designing two graph-based neural network architectures and incorporating them into a joint optimization stage. Furthermore, to explore the information propagation problem with dynamic environment in a large-scale market graph, we extend the GME model with parallelizing competitiveness quantification and hierarchical propagation algorithm. Finally, we conduct extensive experiments on real-world data. The experimental results clearly demonstrate the effectiveness of our proposed model. [ABSTRACT FROM AUTHOR]

• We analyze the basic solution of a generalized differential operator. • We give a class of convolution kernel function. • We propose a semi-supervised multi-layer convolution kernel SVM algorithm. • We define two semi-supervised methods: SSMCK-MKL and SSMCK-AO. In many practical credit evaluation problems, a lot of manpower as well as financial and material resources are required to label samples. Therefore, in the process of labeling, only a small number of samples with category labels can be obtained to train classification models and a large number of customer samples is abandoned without category labels. To solve this problem, we introduce a semi-supervised support vector machine (SVM) technology and combines it with a multi-layer convolution kernel to construct a semi-supervised multi-layer convolution kernel SVM (SSMCK) for category customer credit assessment data sets. We first use a basic solution of the generalized differential operator to generate a base convolution kernel function in the H 1 space, and then use the multi-layer strategy of deep learning to construct the multi-layer convolution kernel in the H 2 and H 3 space (called the family of multi-layer convolution kernel) by using the kernel functions in the H 1 space. We further propose a semi-supervised multi-layer convolution kernel SVM algorithm based on the category center estimation and develop two novel SSMCK methods to improve the classification ability: the SSMCK based on multi-kernel learning (SSMCK-MKL) and the SSMCK based on alternative optimization (SSMCK-AO). Finally, experimental verification and analysis is carried out on three customer credit evaluation data sets. The results show that our methods outperforms or are comparable to some the state-of-the-art credit evaluation models. [ABSTRACT FROM AUTHOR]

Summary: Fluoride is associated with a decrease in the incidence of dental caries, but excess fluoride can lead to enamel fluorosis, a defect that occurs during tooth enamel formation. In fibroblasts, the Arhgap gene encodes a RhoGAP, which regulates the small G protein designated RhoA. Fluoride treatment of fibroblasts inactivates RhoGAP, thereby activating RhoA, which leads to elevation of filamentous actin (F-actin). Since RhoA is a molecular switch, our hypothesis is that in ameloblasts, fluoride may alter the cytoskeleton through interference with the Rho signaling pathway. Our objective was to measure the effects of sodium fluoride on F-actin using tooth organ culture and confocal microscopy. The results indicated that cellular responses to fluoride include elevation of F-actin in ameloblasts. It was concluded from immunohistochemistry, RT-PCR and confocal approaches that the components of the Rho pathway are present in ameloblasts, and that the response to fluoride involves the Rho/ROCK pathway. [Copyright &y& Elsevier]

• We propose a standardized symmetric regularity criterion. • We define a novel structure of initial model of GMDH. • We use probabilistic SVM as base learner and integrate the SVM with GMDH. • We design a special classifier ensemble selection approach named GMDH-PSVM. • Our method may obtain good classification results. Support vector machine (SVM) provides a good classification and regression ability, especially, for small sample learning. However, in practice, the learning ability of implemented SVM is occasionally far from the expected level. Group method of data handling neural network (GMDH-NN) has been applied in various fields for pattern recognition and data mining. It makes it possible to automatically find interrelations in data, to select an optimal structure of network or model and to improve the accuracy of existing algorithms. In this work we propose to take the advantages of GMDH-NN for further increasing the classification performance of SVM. One weakness of the symmetric regularity criterion of GMDH-NN is that if one of the input attributes has a relatively big range, then it may overcome the other attributes. Thus, we first define a standardized symmetric regularity criterion (SSRC) to evaluate and select the candidate models, and optimize a classifier ensemble selection approach. Secondly, we define a novel structure of initial model of GMDH-NN which is from the posterior probability outputs of SVMs. These probabilistic outputs are generated from the improved Platt's probabilistic outputs. Thirdly, in real classification tasks, different classifiers usually have different classification advantages. So we use probabilistic SVM as base learner and integrate the probabilistic SVMs with GMDH-NN, and then propose a special classifier ensemble selection approach for probabilistic SVM classifiers based on GMDH-NN called GMDH-PSVM. Moreover, we use the Borda sorting and Random weighted Borda sorting to discuss the results of our experiments. Experiments on standard UCI datasets demonstrate the effectiveness of our method. [ABSTRACT FROM AUTHOR]