We explore the possibility of meta-learning for the language-independent unsupervised tokenization problem for English, Russian, and Chinese. We implement the meta-learning approach for automatic determination of hyper-parameters of the unsupervised tokenization model proposed in earlier works, relying on various human-independent fitness functions such as normalised anti-entropy, compression factor and cross-split F1 score, as well as additive and multiplicative composite combinations of the three metrics, testing them against the conventional F1 tokenization score. We find a fairly good correlation between the latter and the additive combination of the former three metrics for English and Russian. In case of Chinese, we find a significant correlation between the F 1 score and the compression factor. Our results suggest the possibility of robust unsupervised tokenization of low-resource and dead languages and allow us to think about human languages in terms of the evolution of efficient symbolic communication codes with different structural optimisation schemes that have evolved in different human cultures.
The fast multipole boundary element method finds applications in computer simulation used for a wide variety of electromagnetic and acoustic phenomena. FMM is a tool for the boundary element method optimization. The paper proposes a block-matrix representation of multipole operators for the fast multipole boundary element method. A modification of the FMM is proposed using such a representation of the multipole operators. One of the advantages of the offered method is that the system of linear algebraic equations occurring after its use can be solved by direct methods. Moreover, it is possible to use non-specialized preconditioners in iterative methods to solve SLAE. The matrix assembled in the offered method can be used to precondition the SLAE, obtained by the fast multipole boundary element method, without the proposed modification. The article describes an algorithm for a simple method to obtain the matrix form of multipole operators. A comparison of the fast multipole boundary element method without offered modification and with block-diagonal preconditioning and with the preconditioning proposed in the article is given. During the comparison, harmonics of different orders were used to construct the preconditioner. The use of the proposed method as a preconditioner makes the SLAE preconditioning for the FMM without offered modification of boundary elements relatively simple and sufficiently high quality, which accelerates the convergence of iterative methods.
An approach to modeling multiphase flow of a gas-liquid compressible mixture taking into account the liberation and dissolution of gas is considered. Approach is based on an implicit pressure calculation using the finite element method and explicit phase transport. A computational scheme for modeling the processes of gas component transfer between oil and gas phases based on the known pressure dependence of the gas factor is described. In the considered approach the changes of phases volume appearing during the flow process are taken into account in the right-hand side of the equation used for pressure calculation. An iterative method for solving the obtained nonlinear boundary value problem for pressure calculation is described. This method based on the linearization of the pressure dependence of phases volume changes in the right-hand sides of the equation and boundary conditions. The verification of the proposed method is carried out on a test problem from the comparative project SPE-1. At each time step it takes an average of two nonlinearity iterations. The resulting plots of well data and gas phase distribution data are in good agreement with the results of the SPE-1 project participants.
The problem of optimal management of flight schedules and aircraft assignments is considered. Optimal regulation of schedules implies creating or modifying them in such a way as to minimize system losses due to ongoing disruptions. The total deviation of the corrected schedule from the specified aircraft departure schedules was used as a loss estimate. It is shown that the described technological system belongs to parallel-sequential systems. Accordingly, the considered problem of system control is NP-hard and has no effective algorithms for an exact solution. The article provides a short overview of existing approaches to solving the problem of schedule management and its modifications. The original formal statement of the problem is given, the decomposition of the problem is substantiated, and an algorithm for its approximate solution is presented. An illustrative example is given, and the statistics of testing software implementations of the decomposition algorithm and solving the problem of schedule control on real data are presented, proving the actual effectiveness of the developed toolkit.
Image processing is widely used in many fields, including electronics, computer science, energy, construction, medicine, and self-driving cars. Deep learning based object detection methods have also evolved to detect in real-time. Depending on the nature of the object or image to be processed, the performance of image processing methods will vary from application to application. Therefore, researchers test different types of algorithms in each case and compare performance results. The study aims to detect and report selected objects, such as insulators and vibration dampers. This video was recorded by using an UAV in a power transmission line. The research aims to identify the insulator from the input data processed by image processing, create an image database, and then report the visible physical damages. This study summarizes the results of a study in which video insulators and vibration dampers were detected using the DL youonly look once algorithm.
