The work is aimed to study relationship between vessel endothelial growth factor (VEGF) serum levels and hypoxia in patients with severe neurological disorders. Overviewed literature sources indicate activation of VEGF synthesis during hypoxia. But standard correlation analysis does not reveal a statistically significant association between VEGF levels and pulse oxymetry parameters objectively assessing oxygen supply. Due to a proposed in the paper technique it became possible to reveal valid relationship between VEGF levels and oxygenation. This relationship can be described as increase of correlation between VEGF and immunological complement C4 when oxygenation level is below a certain threshold. Developed method includes searching of such boundaries for oxymetry parameters that correlation coefficients between VEGF and some additional factor Z in formed group are maximally different. ranges. The original technique based on permutation test was used not only to asses validity of effects associated with each oxymetry parameter but also to evaluate joint effect by full group of such factors. Great number of variable from database were tested as additional factors. So additional correction of validity was made to take into account multiple testing.

Reduction of the number of parameters is one of the most important goals in Deep Learning. In this article we propose an adaptation of Doubly Stochastic Variational Inference for Automatic Relevance Determination (DSVI-ARD) for neural networks compression. We find this method to be especially useful in language modeling tasks, where large number of parameters in the input and output layers is often excessive. We also show that DSVI-ARD can be applied together with encoder-decoder weight tying allowing to achieve even better sparsity and performance. Our experiments demonstrate that more than 90% of the weights in both encoder and decoder layers can be removed with a minimal quality loss. * These two authors contributed equally; the ordering of their names was chosen arbitrarily. The work was done when the first author was an intern at the Samsung R&D Institute.

No abstract

Целью работы является исследование связи фактора роста эндотелия сосудов VEGF в сыворотке крови c гипоксией в группах пациентов, страдающих тяжёлыми неврологическими заболеваниями. Имеющиеся в литературе данные свидетельствуют об активации синтеза VEGF при гипоксии. Вместе с тем стандартный корреляционный анализ не позволил достоверно выявить по клиническим данным наличие связи между уровнем VEGF и параметрами оксиметрии, объективно характеризующими снабжение организма кислородом. В статье представлена методика, позволившая статистически достоверно доказать существование указанной связи, сводящейся к увеличению корреляции между VEGF и комплементом C4, а также между VEGF и белками S100 при снижении уровня оксигенации ниже некоторого порогового значения. Методика основывается на построении оптимальных разбиений интервалов значений показателей оксиметрии. Использовалась оригинальная методика статистической верификации, основанная на использовании перестановочных тестов, которая позволила совместно учесть отдельные эффекты по группе показателей оксиметрии, а также провести коррекцию значимости с целью учёта множественного тестирования. Ключевые слова: фактор роста эндотелия сосудов, гипоксия, верификация, перестановочный тест, множественное тестированиеThe work is aimed to study relationship between vessel endothelial growth factor (VEGF) serum levels and hypoxia in patients with severe neurological disorders. Overviewed literature sources indicate activation of VEGF synthesis during hypoxia. But standard correlation analysis does not reveal a statistically significant association between VEGF levels and pulse oxymetry parameters objectively assessing oxygen supply. Due to a proposed in the paper technique it became possible to reveal valid relationship between VEGF levels and oxygenation. This relationship can be described as increase of correlation between VEGF and immunological complement C4 when oxygenation level is below a certain threshold. The method is based on optimal partitioning of pulse oxymetry parameters ranges. The original technique based on permutation test was used to asses validity of joint effect by group of oxymetry parameters. This technique was applied for validity correction aimed to take into account multiple testing.

The paper considers a new method for finding correlations distorted by the influence of a third factor. In other words, the method is designed to search for effects consisted in the existence a significant correlation between two variables in a group of observations received from the main sample by removal of the observations with extremal values of a third factor. Testing for such effects should include both an assessment of the statistical significance of the correlation in the subgroup and the significance of the influence of the third factor. Potentially, this can be done using the previously developed method of valid conditional linear regularities (VCLR). The statistics used in the VCLR method is the maximum of the functional, which depends on the correlation coefficients in the subgroups identified using the threshold for the third factor and the sizes of these subgroups. The disadvantage of this method is that it cannot be used if the maximum of the functional corresponds to a threshold value that cuts off a small group. This drawback did not allow to adequately assess the significance of the effect associated with the existence of a significant negative correlation between the serum levels of antibodies to factor VIII and NR2 in the group of patients with ischemic stroke and transient ischemic attack after excluding patients with abnormally high levels of vascular endothelial growth factor. An alternative method was proposed, which is based on a permutation test. At that the statistics of the test is a minimum p-value from those characterizing the correlation coefficients calculated using the normal approximation of the Fisher's z-transform corresponding to all possible threshold values for the third factor. The use of the new criterion made it possible to adequately assess the significance of the observed effect.