Meemi: A simple method for post-processing and integrating cross-lingual word embeddings

Word embeddings have become a standard resource in the toolset of any Natural Language Processing practitioner. While monolingual word embeddings encode information about words in the context of a particular language, cross-lingual embeddings define a multilingual space where word embeddings from two or more languages are integrated together. Current state-of-the-art approaches learn these embeddings by aligning two disjoint monolingual vector spaces through an orthogonal transformation which preserves the structure of the monolingual counterparts. In this work, we propose to apply an additional transformation after this initial alignment step, which aims to bring the vector representations of a given word and its translations closer to their average. Since this additional transformation is non-orthogonal, it also affects the structure of the monolingual spaces. We show that our approach both improves the integration of the monolingual spaces and the quality of the monolingual spaces themselves. Furthermore, because our transformation can be applied to an arbitrary number of languages, we are able to effectively obtain a truly multilingual space. The resulting (monolingual and multilingual) spaces show consistent gains over the current state-of-the-art in standard intrinsic tasks, namely dictionary induction and word similarity, as well as in extrinsic tasks such as cross-lingual hypernym discovery and cross-lingual natural language inference.

Buckling Pattern Transition Of Periodic Porous Elastomers Induced By Proportional Loading Conditions

This paper focuses on the buckling instabilities of periodic porous elastomers under combined multiaxial loading. A numerical model based on the periodic boundary condition (PBC) for the 2D representative volume element (RVE) is proposed, in which two proportional loading parameters are employed to control the complex stressing state applied to the RVE model. A homogenization-based orthogonal transformation matrix is established by satisfying the equality of the total work rate to realize a proportional multiaxial loading on the RVE. First, the transition behavior of buckling patterns of periodic porous structures is revealed through instability analysis for the RVE consisting of [Formula: see text] primitive cells with circular holes subjected to different proportional loading conditions. Simulation results show that the first-order buckling mode of RVE may change suddenly from a uniaxial shearing buckling pattern to a biaxial rotating buckling pattern at a critical loading proportion. Then the influences of the number of primitive cells in the enlarged RVE on the buckling behavior are discussed. When the number of primitive cells in any enlarging direction is odd, the points of buckling pattern transition of the enlarged RVEs vary significantly with the number of cells in RVE. When the number of primitive cells is even in both enlarging directions, there is no apparent difference for the critical buckling stresses of the enlarged RVEs.

Image compression based on classification coding of constant-pitched functions transformers

The subject of research in the article are the processes of video image processing using an orthogonal transformation for data transmission in information and telecommunication networks. The aim is to build a method of compression of video images while maintaining the efficiency of its delivery at a given informative probability. That will allow to provide a gain in the time of delivery of compressed video images, a necessary level of availability and authenticity at transfer of video data with preservation of strictly statistical regulations and the controlled loss of quality. Task: to study the known algorithms for selective processing of static video at the stage of approximation and statistical coding of the data based on JPEG-platform. The methods used are algorithm based on JPEG-platform, methods of approximation by orthogonal transformation of information blocks, arithmetic coding. It is a solution of scientific task-developed methods for reducing the computational complexity of transformations (compression and decompression) of static video images in the equipment for processing visual information signals, which will increase the efficiency of information delivery.The following results were obtained. The method of video image compression with preservation of the efficiency of its delivery at the set informative probability is developed. That will allow to fulfill the set requirements at the preservation of structural-statistical economy, providing a gain in time to bring compressed images based on the developed method, relative to known methods, on average up to 2 times. This gain is because with a slight difference in the compression ratio of highly saturated images compared to the JPEG-2000 method, for the developed method, the processing time will be less by at least 34%.Moreover, with the increase in the volume of transmitted images and the data transmission speed in the communication channel - the gain in the time of delivery for the developed method will increase. Here, the loss of quality of the compressed/restored image does not exceed 2% by RMS, or not worse than 45 dB by PSNR. What is unnoticeable to the human eye.Conclusions. The scientific novelty of the obtained results is as follows: for the first time the method of classification (separate) coding (compression) of high-frequency and low-frequency components of Walsh transformants of video images is offered and investigated, which allows to consider their different dynamic range and statistical redundancy reduced using arithmetic coding. This method will allow to ensure the necessary level of availability and authenticity when transmitting video data, while maintaining strict statistical statistics.Note that the proposed method fulfills the set tasks to increase the efficiency of information delivery. Simultaneously, the method for reducing the time complexity of the conversion of highly saturated video images using their representation by the transformants of the discrete Walsh transformation was further developed. It is substantiated that the perspective direction of improvement of methods of image compression is the application of orthogonal transformations on the basis of integer piecewise-constant functions, and methods of integer arithmetic coding of values of transformant transformations.It is substantiated that the joint use of Walsh transformation and arithmetic coding, which reduces the time of compression and recovery of images; reduces additional statistical redundancy. To further increase the degree of compression, a classification coding of low-frequency and high-frequency components of Walsh transformants is developed. It is shown that an additional reduction in statistical redundancy in the arrays of low-frequency components of Walsh transformants is achieved due to their difference in representation. Recommendations for the parameters of the compression method for which the lowest value of the total time of information delivery is provided are substantiated.

Automatic Face Emotion Recognition With the Aid of Probability-Based Bird Swarm-Trained Neural Network

This paper intends to develop a novel FER model, which consists of four stages: (1) face detection, (2) feature extraction, (3) dimension reduction, and (4) classification. In this context, the face detection is done using Viola Jones method (VJ). It is the first object recognition model to offer better recognition rates in real-time. Further, features extraction techniques like local binary pattern (LBP) and discrete wavelet transform (DWT) are used for extracting the features from face detected images. Moreover, the dimension reduction of features is done using principal component analysis (PCA), which is an arithmetical process that exploits an orthogonal transformation to exchange a group of annotations of probably interrelated constraints. The classification procedure is performed using neural network (NN), with the new training algorithm called bird swarm algorithm, which is modified based on probability and hence termed as probability-based BSA (P-BSA).

463Association of physical activity with obstructive sleep opena

Background Obstructive sleep apnea (OSA) is the most common sleep disorder in clinical practice, which leads to impaired quality of life and is associated with adverse health outcomes. Due to limitations of continuous positive airway pressure (CPAP), the gold standard treatment for OSA, studies have recommended lifestyle interventions such as physical activity to prevent OSA. This study was to investigate the association of physical activity with OSA risk among adult Chinese. Methods 9733 participants aged 35-74 years were selected from baseline survey of the Guangzhou Heart Study. OSA was ascertained by using Berlin Questionnaire and the physical activity, including leisure-time physical activity (LTPA), occupational activity and transport activity, was measured with modified Global Physical Activity Questionnaire. Principal component analysis was used to extract the patterns of LTPA with varimax orthogonal transformation. Odds ratio (OR) with 95% confidence interval (95% CI) was calculated by using logistic regression method. Results For all participants, LTPA (High vs. Inactive, OR: 0.81, 95% CI: 0.64 - 1.03), occupational activity (Vigorous vs. Retirement, OR:1.28, 95% CI: 0.93 - 1.75) and transport activity (High vs. Retirement, OR: 1.05, 95% CI: 0.69 - 1.60) were not associated with OSA risk after considering potential confounders. Any specific component of LTPA and two LTPA patterns were not associated with OSA risk, either. Stratified analysis yielded similar nonsignificant association of OSA risk with three dimensions of physical activity in both retirement group and non-retirement group. Conclusion This study found that three dimensions of physical activity, including LTPA, transport activity and occupational activity, were not associated with any risk of OSA. Future studies with longitudinal design are needed. Key message Physical activity may not decrease the risk of obstructive sleep apena.

Three-Dimensional Weighted Multiple-Relaxation-Time Pseudopotential Lattice Boltzmann Method for Multiphase Flow

The pseudopotential lattice Boltzmann (LB) method has been widely used for simulating multiphase flow due to its concise concept and computational simplicity. In this paper, based on the weighted orthogonal transformation matrix, a three-dimensional (3D) weighted multiple-relaxation-time pseudopotential lattice Boltzmann method (WRMT-LBM) is developed, in which the standard lattice stencil D3Q19 is adopted. Compared with the classical multiple-relaxation-time pseudopotential lattice Boltzmann method (CMRT-LBM) based on the orthogonal transformation matrix, the expressions of the equilibrium density distribution function and discrete force term in moment space are simplified in the present model, which contributes to simplifying the program implementation and improving the computational efficiency. Moreover, an additional discrete source term in moment space compatible with the proposed model is introduced to achieve tunable surface tension. A series of numerical tests are then implemented to investigate the performance of the proposed model. Compared with the CMRT-LBM, the results of the present model can achieve lower spurious velocity and higher computational efficiency while keeping comparable accuracy. Furthermore, using the present model, three benchmark cases, including droplet oscillation, droplet impacting on wall and droplet impact on thin film, are performed to investigate the performance of this model. The numerical results are in good agreement with the analytical solutions or the empirical correlations in the literature, which demonstrates that the present model can simulate the multiphase flow with large density ratio.

Optimization of computational complexity of an artificial neural network

The article deals with the modelling of Artificial Neural Networks as an information transmission system to optimize their computational complexity. The analysis of existing theoretical approaches to optimizing the structure and training of neural networks is carried out. In the process of constructing the model, the well-known problem of isolating a deterministic signal on the background of noise and adapting it to solving the problem of assigning an input implementation to a certain cluster is considered. A layer of neurons is considered as an information transformer with a kernel for solving a certain class of problems: orthogonal transformation, matched filtering, and nonlinear transformation for recognizing the input implementation with a given accuracy. Based on the analysis of the proposed model, it is concluded that it is possible to reduce the number of neurons in the layers of neural network and to reduce the number of features for training the classifier.

Errors of Indices in Household Surveys of Punjab Urban through Principal Components

In this research article, household wealth indices are calculated to estimate the sampling errors, which gave us complete information on the quality and reliability of published data upon the household surveys. Estimates are calculated based on simple random sampling, which contains sampling errors. Here principal components analysis (PCA) estimate standard errors of wealth indices as orthogonal transformation to develop solid measures of individual economic status. These measures evaluate the significance and explain the living status and economic dissimilarity of Punjab urban. Instrumental variables are used here to the enlightened social status of the Punjab urban area of Pakistan by using PCA of household surveys. Twenty-five variables are included in this study. Total variance analysis explains the variation of total components. A comparison study (PCA) approaches to estimating the standard error of indices of the household survey are presented in this paper. We conclude that errors of indices in household surveys through PCA, when compared to direct measures of estimating household wealth indices, are an efficient and reliable method.