We revisit the idea of brain damage, i.e. the pruning of the coefficients of a neural network, and suggest how brain damage can be modified and used to speedup convolutional layers in ConvNets. The approach uses the fact that many efficient implementations reduce generalized convolutions to matrix multiplications. The suggested brain damage process prunes the convolutional kernel tensor in a group-wise fashion. After such pruning, convolutions can be reduced to multiplications of thinned dense matrices, which leads to speedup. We investigate different ways to add group-wise prunning to the learning process, and show that severalfold speedups of convolutional layers can be attained using group-sparsity regularizers. Our approach can adjust the shapes of the receptive fields in the convolutional layers, and even prune excessive feature maps from ConvNets, all in data-driven way.
a) Data availability (b) Satellite imagery (c) GFS model (d) Precipitation detection Figure 1: (a) The availability of input data: full field of view of the Meteosat-8 satellite, the currently processed area inside it and the coverage of Roshydromet radars. (b) IR-097 (infrared channel) from the Meteosat-8 satellite imagery. (c) Total cloud water (cloud liquid water + cloud ice) from the GFS model of the atmosphere. (d) Our reconstruction of the precipitation field.ABSTRACT Precipitation nowcasting is a short-range forecast of rain/snow (up to 2 hours), often displayed on top of the geographical map by the weather service. Modern precipitation nowcasting algorithms rely on the extrapolation of observations by ground-based radars via optical flow techniques or neural network models. Dependent on these radars, typical nowcasting is limited to the regions around their locations. We have developed a method for precipitation nowcasting based on geostationary satellite imagery and incorporated the resulting data into the Yandex.Weather precipitation map (including an alerting service with push notifications for products in the Yandex ecosystem), thus expanding its coverage and paving the way to a truly global nowcasting service.
Presently, there is no generally accepted approach to automated measurement of pulse duration in cardiac cycles. This makes it difficult to compare the results obtained by different cardiograph models. At present, this problem is not very urgent because the duration of cardiopulses is still measured manually from the ECG records with an accuracy of about 10 msec [8, 13]. However, methods of comparison of the results of repeated examinations of the same patient during a significant length of time that are presently being introduced used into medical practice require the accuracy of the measurements to be at least 1 msec. Electrocardiographs providing such accuracy of measurements are available, but differences in algorithms of automated measurement in this case becomes a significant problem.The goal of this work was to consider various algorithms of measurement of duration, or, more precisely, various algorithms of determining the beginning and end of cardiopulses, which allows the duration of the pulse to be easily calculated. Development of such algorithms is one of the tasks of mathematical statistics. A comparative survey of algorithms given below treats this problem from the point of view of electrocardiography. Noise stability and accuracy of determination of the beginning of a pulse are chosen as the criteria for comparison of various algorithms. Specific Physiological Features of the Be~nning of Cardiac Cycle PulsesRecent studies show [20] that the shape of cardiopulses at their beginning is rather complicated and has various phases. Each cardiopulse (especially the P wave) is preceded with a fluctuation process of spontaneous generation of artifacts (background myographic noise). At a certain moment in time, sinoatrial node ceils undergo collective excitation which is transmitted to adjacen t regions of the myocardial auricles. The steepness and shape of the leading edge of the excitation pulse are determined by the rate of excitation transmission and changes in the total volume of depolarized ceils. Initially, the process propagates in three dimensions, so it can be concluded that its rate is proportional to the third power of time. The measured rate is proportional to the third power of time because of the vector selectivity of ECG leads. Taking into account the thickness of myocardial walls (several mm in the region of the auricles), it can be concluded that the first steep phase of the process occurs within 1-4 msec. It is followed by a more flattened phase and the end of the pulse. The shape of P wave at its beginning is determined by the geometry of auricle muscles, anisotropy of excitation transmission rates, and vector selectivity of the lead.The process of appearance of the QRS complex is different from that described above. The QRS complex is preceded by the repolarization process in auricles and signals from the atrioventricular node, which transmit excitation with a delay to the His' bundle. A branched network of the His' bundle conducting paths distributes excitation over the endocardial pap...
, Санкт-Петеpбуpгский Балтийский госудаpственный технический унивеpситет им. Д. Ф. Устинова "Военмех" Модульная синхpонная индуктоpная машина в системе электpопpивода 1. Констpукция модульной синхpонной индуктоpной машины
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.