Papers by Andrey Gritsenko
Modern Applied Science, Mar 25, 2015
The aim of this paper is to provide a description of machine learning based scheduling approach f... more The aim of this paper is to provide a description of machine learning based scheduling approach for high-loaded distributed systems that have patterns of tasks/queries that occur recurrently in workflow. The core of this approach is to predict the future workflow of the system depending on previous tasks/queries using supervised learning. First of all, the workflow is analyzed using hierarchical clustering to reveal sets of tasks/queries. Revealed sets of tasks/queries then undergo restructuring to represent patterns of recurrent tasks/queries. Later these patterns become the object of the forecasting process performed using neural network. Information on predicted tasks/queries is used by the resource management system (RMS) to perform efficient schedule. To estimate the performance of the described method it was at first realized as a module of the simulation tool Alea that models the work of high-performance distributed systems and then compared with other state-of-the-art scheduling algorithms. The simulation was produced for two datasets: in one of the experiments the proposed method showed best results, and in the other it was inferior to just a single method, though it was much better than commonly used standard scheduling algorithms.
Extreme Learning Machines for VISualization+R: Mastering Visualization with Target Variables
Cognitive Computation, Dec 22, 2017
The current paper presents an improvement of the Extreme Learning Machines for VISualization (ELM... more The current paper presents an improvement of the Extreme Learning Machines for VISualization (ELMVIS+) nonlinear dimensionality reduction method. In this improved method, called ELMVIS+R, it is proposed to apply the originally unsupervised ELMVIS+ method for the regression problems, using target values to improve visualization results. It has been shown in previous work that the approach of adding supervised component for classification problems indeed allows to obtain better visualization results. To verify this assumption for regression problems, a set of experiments on several different datasets was performed. The newly proposed method was compared to the ELMVIS+ method and, in most cases, outperformed the original algorithm. Results, presented in this article, prove the general idea that using supervised components (target values) with nonlinear dimensionality reduction method like ELMVIS+ can improve both visual properties and overall accuracy.
Deformable Surface Registration with Extreme Learning Machines
Proceedings in adaptation, learning and optimization, Oct 17, 2018
One of the most important open problems in the field of computer-aided design and computer graphi... more One of the most important open problems in the field of computer-aided design and computer graphics is the task of surface registration for non-isometric cases. One of the approaches of addressing surface registration problem is to find the point-wise correspondence between surfaces using state-of-the-art shape descriptors. This paper introduces an improvement to this approach by means of Extreme Learning Machines. The ELM model is trained to distinguish pairs of corresponding points from non-corresponding ones on the dataset with highly non-isometric distortions between models. The proposed method is compared with original shape descriptors. The results show the increase of accuracy in surface registration task, and also reveal the bottleneck of the state-of-the-art shape descriptors.
Graph transfer learning
Knowledge and Information Systems, Dec 21, 2022
A Comparison Model of Scheduling and Resource Distribution Algorithms for Large Scale Computing Systems
V mire naučnyh otkrytij, Nov 30, 2014
Проблема планирования, заключающаяся в выполнении определенных заданий на указанных ресурсах в ук... more Проблема планирования, заключающаяся в выполнении определенных заданий на указанных ресурсах в указанный момент времени, лежит в сердце любой распределенной высокопроизводительной вычислительной системы. Разработка алгоритмов планирования обычно основана на стремлении улучшить производительность по различным показателям, тем самым осложняя процесс сравнения таких алгоритмов. Такое осложнение объясняется тем, что целенаправленно улучшая результаты по одному показателю производительности, алгоритмы показывают плохие результаты по другим показателям. Чтобы справиться с возникающей проблемой была предложена унифицированная модель сравнения. Прежде всего, из большого числа используемых в настоящее время показателей эффективности были выбраны несколько таким образом, чтобы сформировать набор, позволяющий всесторонне оценить производительность алгоритмов планирования с различных точек зрения. Затем описывается модель сравнения, в основе которой лежат два различных метода принятия решений. Наконец, даются общие рекомендации по использованию предложенной модели сравнения.
arXiv (Cornell University), Oct 6, 2013
The aim of this paper is to provide a description of deep-learning-based scheduling approach for ... more The aim of this paper is to provide a description of deep-learning-based scheduling approach for academicpurpose high-performance computing systems. Academicpurpose distributed computing systems' (DCS) share reaches 17.4% amongst TOP500 supercomputer sites (15.6% in performance scale) that make them a valuable object of research. The core of this approach is to predict the future workflow of the system depending on the previously submitted tasks using deep learning algorithm. Information on predicted tasks is used by the resource management system (RMS) to perform efficient schedule.
Indian Journal of Science and Technology, 2016
This article describes the integration method of mobile applications with corporate information s... more This article describes the integration method of mobile applications with corporate information system (CIS) through the previously proposed interaction architecture based on integration environment. The structure of the integration environment, which includes new approaches to the organization of data storage, conversion and presentation, is described in detail. Also this article represents the algorithm for implementation of CIS mobile applications integration including the proposed structure of integration environment.
Graph transfer learning
Knowledge and Information Systems
2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI), 2020
Twin study is one of the major parts of human brain research that reveals the importance of envir... more Twin study is one of the major parts of human brain research that reveals the importance of environmental and genetic influences on different aspects of brain behavior and disorders. Accurate characterization of identical and fraternal twins allows us to infer on the genetic influence in a population. In this paper, we propose a novel pair-wise classification pipeline to identify the zygosity of twin pairs using the resting state functional magnetic resonance images (rs-fMRI). The new feature representation is utilized to efficiently construct brain network for each subject. Specifically, we project the fMRI signal to a set of cosine series basis and use the projection coefficients as the compact and discriminative feature representation of noisy fMRI. The pair-wise relation is encoded by a set of twin-wise correlations between functional brain networks across brain regions. We further employ hill climbing variable selection to identify the most genetically affected brain regions. T...
Deformable Surface Registration with Extreme Learning Machines
Proceedings in Adaptation, Learning and Optimization, 2018
One of the most important open problems in the field of computer-aided design and computer graphi... more One of the most important open problems in the field of computer-aided design and computer graphics is the task of surface registration for non-isometric cases. One of the approaches of addressing surface registration problem is to find the point-wise correspondence between surfaces using state-of-the-art shape descriptors. This paper introduces an improvement to this approach by means of Extreme Learning Machines. The ELM model is trained to distinguish pairs of corresponding points from non-corresponding ones on the dataset with highly non-isometric distortions between models. The proposed method is compared with original shape descriptors. The results show the increase of accuracy in surface registration task, and also reveal the bottleneck of the state-of-the-art shape descriptors.
The aim of this paper is to provide a description of deep-learning-based scheduling approach for ... more The aim of this paper is to provide a description of deep-learning-based scheduling approach for academicpurpose high-performance computing systems. Academicpurpose distributed computing systems' (DCS) share reaches 17.4% amongst TOP500 supercomputer sites (15.6% in performance scale) that make them a valuable object of research. The core of this approach is to predict the future workflow of the system depending on the previously submitted tasks using deep learning algorithm. Information on predicted tasks is used by the resource management system (RMS) to perform efficient schedule.
Lecture Notes in Computer Science, 2012
The use of general descriptive names, registered names, trademarks, etc. in this publication does... more The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use.
Lecture Notes in Computer Science, 2015
This paper presents an extension of the well-known Extreme Learning Machines (ELMs). The main goa... more This paper presents an extension of the well-known Extreme Learning Machines (ELMs). The main goal is to provide probabilities as outputs for Multiclass Classification problems. Such information is more useful in practice than traditional crisp classification outputs. In summary, Gaussian Mixture Models are used as post-processing of ELMs. In that context, the proposed global methodology is keeping the advantages of ELMs (low computational time and state of the art performances) and the ability of Gaussian Mixture Models to deal with probabilities. The methodology is tested on 3 toy examples and 3 real datasets. As a result, the global performances of ELMs are slightly improved and the probability outputs are seen to be accurate and useful in practice.
Solve Classification Tasks with Probabilities. Statistically-Modeled Outputs
In this paper, an approach for probability-based class prediction is presented. This approach is ... more In this paper, an approach for probability-based class prediction is presented. This approach is based on a combination of a newly proposed Histogram Probability (HP) method and any classification algorithm (in this paper results for combination with Extreme Learning Machines (ELM) and Support Vector Machines (SVM) are presented). Extreme Learning Machines is a method of training a single-hidden layer neural network. The paper contains detailed description and analysis of the HP method by the example of the Iris dataset. Eight datasets, four of which represent computer vision classification problem and are derived from Caltech-256 image database, are used to compare HP method with another probability-output classifier [11, 18].
Extreme Learning Machines for VISualization+R: Mastering Visualization with Target Variables
Cognitive Computation
The current paper presents an improvement of the Extreme Learning Machines for VISualization (ELM... more The current paper presents an improvement of the Extreme Learning Machines for VISualization (ELMVIS+) nonlinear dimensionality reduction method. In this improved method, called ELMVIS+R, it is proposed to apply the originally unsupervised ELMVIS+ method for the regression problems, using target values to improve visualization results. It has been shown in previous work that the approach of adding supervised component for classification problems indeed allows to obtain better visualization results. To verify this assumption for regression problems, a set of experiments on several different datasets was performed. The newly proposed method was compared to the ELMVIS+ method and, in most cases, outperformed the original algorithm. Results, presented in this article, prove the general idea that using supervised components (target values) with nonlinear dimensionality reduction method like ELMVIS+ can improve both visual properties and overall accuracy.
A key strength of twin studies arises from the fact that there are two types of twins, monozygoti... more A key strength of twin studies arises from the fact that there are two types of twins, monozygotic and dizygotic, that share differing amounts of genetic information. Accurate differentiation of twin types allows efficient inference on genetic influences in a population. However, identification of zygosity is often prone to errors without genotying. In this study, we propose a novel pairwise feature representation to classify the zygosity of twin pairs of resting state functional magnetic resonance images (rs-fMRI). For this, we project an fMRI signal to a set of basis functions and use the projection coefficients as the compact and discriminative feature representation of noisy fMRI. We encode the relationship between twins as the correlation between the new feature representations across brain regions. We employ hill climbing variable selection to identify brain regions that are the most genetically affected. The proposed framework was applied to 208 twin pairs and achieved 94.19%...
A Comparison Model of Scheduling and Resource Distribution Algorithms for Large Scale Computing Systems
V mire nauchnykh otkrytiy, 2014
Probabilistic Methods for Multiclass Classification Problems
Proceedings in Adaptation, Learning and Optimization, 2016
Lecture Notes in Computer Science, 2015
This paper presents an extension of the well-known Extreme Learning Machines (ELMs). The main goa... more This paper presents an extension of the well-known Extreme Learning Machines (ELMs). The main goal is to provide probabilities as outputs for Multiclass Classification problems. Such information is more useful in practice than traditional crisp classification outputs. In summary, Gaussian Mixture Models are used as post-processing of ELMs. In that context, the proposed global methodology is keeping the advantages of ELMs (low computational time and state of the art performances) and the ability of Gaussian Mixture Models to deal with probabilities. The methodology is tested on 3 toy examples and 3 real datasets. As a result, the global performances of ELMs are slightly improved and the probability outputs are seen to be accurate and useful in practice.
Uploads
Papers by Andrey Gritsenko