Science

Below there are scientific articles written by me (and other scientific projects), in order from recent to older.


Topic models interactive visualization technology (2017, B. Sc. Thesis)

Abstract. In this work we study methods of topic models visualization. We consider several ways of visualizing topic models (including temporal and hierarchical). We set a problem of building topic model spectrum - a permutation of its topics, in which semantically close topics are located nearby. We suggest several algorithms to solve this problem and design a method to assess their quality. The task is generalized for the case of hierarchical topic models. Also we describe informational system VisARTM, which is designed for automated creation and visualization of topic models.

Keywords: Topic Modeling, Visualization.

Links: paper, presentation, code, website.

Scientific advisor: Konstantin Vorontsov

Defended on 29.06.2017 in CCAS.

03.05.2017 - Report on BigARTM seminar. Presentation.


Hierarchic topic models visualization (2016)

Dmitriy S. Fedoriaka

Course work on course "Creating of exploited models of Machine Learning". Supervisor V. V. Strijov.

Abstract. Hierarchic topic models are good tool for representing big amount of text documents. However, displaying such models on screen is difficult problem itself. This paper discusses problem of hierarchic topic models visualization. It introduces concept of tree visualization with polygons. Also it considers problems of quality measuring and representation of additional information. This article also describes implementation of visualization algorithms with usage of BigARTM topic modeling library.

Keywords: Topic Modeling, Visualization, BigARTM.

Main result: Fancy Topic Visualizer.

Links: paper, presentation, code, sysdoc, video.

Mixtures of models of vector autoregression in the problem of time series forecasting (2016)

Fedoriaka D.S., Neychev R.G., Strijov V.V.

Abstract. In this paper we investigate the problem of short-term time series forecasting. We consider the time series of different scales, which are interconnected and have the property of periodicity. Forecasting problem is reduced to the problem of regression, which is solved by using a linear model. To improve its accuracy we propose to use composition of models. The compositions are built using bagging, random subspace method and boosting algorithm AdaBoost. We also propose heuristic iterative algorithm of model composition based on the idea of ​​clustering algorithm K-means. Using the proposed methods, we forecast energy consumption in Turkey and Poland and spot prices on electricty in Germany.

Keywords: time series, forecasting, autoregression, linear regression, boosting, AdaBoost, bagging, mixture of experts.

Language: Russian.

Unpublished.

Links: paper, paper (mirror), presentation, code.


Computer modelling in application to city traffic (2013)

Fedoriaka D.S., Entin J.A.

Abstract. Model of traffic in a big sity is developed. We consider: 1)one- and bideirected movement, 2)intersections, 3)basic requirements of traffic safety. We have created a decision support system for traffic optimization. For computer realization of model we have created project with usage of object-oriented programming.

Keywords: computer simulation, automobile traffic in bigcities, automobiles, roads, buildings, crosses, transport streams optimization

Language: Ukrainian.

Unpublished.

Links: paper, program, video.


Percolation phenomenon in cellular automata (2012)

Fedoriaka D.S., Entin J.A.

Abstract. In this paper, we investigate synchronous cellular automata, similar to the 'Game of Life'. We study the phenomenon of percolation — abrupt filling of all cells during evolution from random initial state when critical concentration is reached. We give theoretical explanation to this phenomena. Also we study the dependency of borders and critical index of percolation from form of the grid and parameters of the automata.

Keywords: cellular automata, percolation.

Language: Ukrainian.

Unpublished.

Links: paper


Modelling and research of the Game of Life and its modifications on triangle, squared and hexagonal grids (2011)

Fedoriaka D.S., Entin J.A.

Abstract. Algorithms and program are created for analysis of counters colonies evolution on two-dimension lattices filled with rectilinear triangles and hexagons under different initial configurations and survival conditions.

Keywords: computer simulation, counters, counters colonies, counters population, evolution, two-dimension lattices, rectilinear triangles and hexagons, counters initial configurations and survival conditions.

Language: Ukrainian.

Published in "Mathematics. Informatics. Physics." (the scientific magazine of Dnipropetrovsk Liceum of Informational Technology), Vol.2 .

Links: paper, program.


Find me in scientific networks: