Software Project

Low Cost Hosting Free Web Space Dedicated Hosting Windows Hosting Trade Show

Displays GoDaddy Review FrontPage Web Hosting Business Hosting

Search the Web

Vladimir V. Krepets' curriculum vitae

Homepage Resume Family tree Photo Guest Book

Full name:

Vladimir Valentinovich Krepets.

Current residence:

&hidden by security reason&, Chevy Chase, Maryland, USA.

Date and place of birth:

&hidden by security reason&. The city of Novosibirsk, Russian Federation, USSR.

Citizenship:

Russia.

Status in the USA:

Permanent resident (green card).

Marital status:

Married, son 2002 year born, daughter 2009 year born.

Professional position:

2011 - curr: .NET Database Application Developer, Penn State Hershey College of Medicine,

Pennsylvania, USA

.

2007 - 2009: Software Engineer in Data Unlimited International, Inc., Maryland, USA.

2005 - 2007: FreelanceSoftware Developer in Math-On-The-Run, LLC., Maryland, USA.

2004 - 2005: Freelance Statistician in the Research Center of Children's Health, RAMS,

Moscow, Russia.

1999 - 2004: Research Scientist

in the Laboratory of Molecular Graphics & Drug Design of

the Institute of Biomedical Chemistry RAMS, Moscow, Russia.

Postgraduate:

1995-1998 in the Computer Center of the Russian Academy of Science on a specialty

05.13.16 - "Application of mathematical methods, mathematical modeling and computer

facilities in scientific researches". Ph.D. thesis (Physics & Mathematics) on a specialty

05.13.18 - "Mathematical modeling, numerical methods and computer software" entitled "The

investigation of stability of feedforward neural networks at the construction of the

regression models of a biological objects" was presented on Dec 21, 2001. Full list of

publications contains more than 15 papers, the most significant of which are listed on the

bottom of the page.

Higher education, Academic degree:

1988-1994 Faculty of Management & Applied Mathematics of Moscow Institute of Physics &

Technology (Moscow), M.Sc. in Applied Mathematics & Physics.

Selected projects:

Cytochemical evaluation of a human condition - individual calendar (Russian copyright

certificate # 200*******

): Software development, statistical models development. First

release has been done with Turbo Pascal. Current release: Interface has been done with

Borland DELPHI; computing module has been done with C++. Project started in 1993. You may

download the latest release of the software here. Unfortunately it is available only with

russian interface.

Neural Network Constructor: Software development, statistical models development. First

release has been done with Turbo Pascal. Current release: Interface has been done with

Borland DELPHI; computing module has been done with C++. Project started in 1995.

Evaluation of the quality of life index. Statistical models development. Used software:

NNC (own software), SPSS, Systat. Project started in 1997, finished in 1998.

Prediction of binding affinities for protein-ligand complexes. Software development with

Microsoft Visual Basic, statistical models development. Project started in 1999, finished

in 2004.

Evaluation of physical health of schoolchildren. Software development, Statistical models

development. Interface has been done with Borland DELPHI; computing module has been done

with C++. Project started in 2003, finished in 2004.

Math on the run. Self-tutoring game software development with Borland DELPHI. Project

started in 2005.

Developed software:

Cytochemical evaluation of a human condition - individual calendar. (Russian copyright

certificate # 200*******)

Neural Network Constructor (if link does not work you may download NNC.zip here)

Math on the run (www.mathontherun.com)

Software experience:

Programming languages: Borland Delphi, ASP.NET, C#/C++/C, Visual Basic, Pascal, Fortran;

SAS macro language and HyperText Markup Language (HTML) though it is not the programming

language;

Database: MS SQL Server 2000, 2005, 2008 and 2012 (express), MS Access, dBase IV

Statistical software application and tools: SPSS, Systat, SAS for Windows and UNIX

environment, SAS EG

Office applications: MS Word, MS Excel, MS PowerPoint, MS Access, etc.

Current scientific interests:

Mathematical modeling, numerical simulation, statistical analysis, constructing an

artificial neural networks, high-level programming. Applying of all above mentioned for

the solving of the problems in such fields of knowledge as the properties of biological

objects, sociology, medicine or any other fields of knowledge.

of 19 cytochemical parameters that characterize the SDG activity, employing the neural

network method, served as the basis for working out algorithm for the calculation of the

integral cytochemical index, which would correspond to the minimum, moderate and maximum

activity degrees of the inflammatory process, and for a highly reliable division of

patients according to these degrees. Also, the determination of the cytochemical index

makes it possible to enhance the efficacy of the therapy.

2. Belkina N.V., Krepets V.V., Shakin V.V. On stable estimation of the parameters of

feedforward neural networks in dealing with biological objects. // AUTOMATION AND REMOTE

CONTROL Vol. 63: (1), pp. 66-75, MAIK NAUKA/INTERPERIODICA, NEW YORK, 2002.

Abstract. This work is a logical extension of the works of Shurygin over the period from

1994 to 1996 and is devoted to the development of approaches to the stable estimation of

the parameters of regression models. The results obtained earlier are extended to the

cases of a nonlinear regression and a feedforward neural network with one hidden layer.

Theoretical results are confirmed by numerical experiments. The problem of numerical

modeling consisted in the construction of a system for the prediction of a change in the

Gibbs free energy (DG) in the course of the formation of protein-protein and protein-

ligand complexes. For the training set, data on 150 complexes of a various nature are

used, for which there exists an experimental estimate (DG). For independent variables,

different rated values of the physicochemical parameters of data for complexes are used.

3. Krepets V.V., Belkina N.V., Skvortsov V.S., Ivanov A.S. Prediction of binding

affinities for protein-ligand complexes by using non-linear models. - Moscow, Questions of

Medical Chemistry, 2000, #5, pp. 262-274. (In Russian)

Abstract. In this paper we present the creation a network model for prediction of the

free energy changes in protein-ligand complexes. The 150 complexes of different nature

were used as a training set. The computational physics-chemical parameters of these

complexes were used as independent variables. Both classical models of multiple linear

regression and several network models with one hidden layer were created. From later the

best was chosen. Significant improvement was shown for network model prediction quality in

comparison with classical model of multiple linear regression (R2 on training - 0,81 and

0,54; R2 on "leave-one-out" procedure - 0,74 and 0,52 respectively).

4. Krepets V.V. Regularization of linear regression models in various metric spaces. -

Moscow, Science & Technology in Russia, 1999, #4 (34), pp. 10-12. (In Russian)

Abstract. Often researcher unable to completely solve a binary classification task by

using existing statistical packages even in that event if it is known that the solution

exists. One of the possible reasons of such failures is the presence of the large single

deviations of the predicted values from really observable values of a dependent variable.

Such deviations are the result of using Euclidean (square-law) metric in standard

statistical packages. In this work we investigate the models constructed in spaces with

the l4, l8 and lY metrics. We reduce theoretical substantiation of usefulness of

introduction the metric spaces with such metrics. Entering of the similar metrics brings

about imposition of the much greater penalty on the observations, poorly inserted in a

construct model and diminution of a maximum error at prediction. Thus, the exactitude of

prediction of one or another magnitude is increased at a model construction.

5. Shishchenko V.M., Krepets V.V., Petrichuk S.V., Duchova Z.N. The perspectives of

neural networks method use for a prognosis of the first year child development. - Moscow,

Science & Technology in Russia, 1999, #3 (33), pp. 9-11. (In Russian)

Abstract. The program predicting children growth in perinatal, neonatal and early age

periods was developed in the Institute of Pediatrics, Russian Academy of Medical Science.

This program allows making a prognosis of the child growth from birth till two years based

on clinic-cytochemical data of the father and mother before pregnancy. The system of

medical measures directing to eliminate of unfavorable prognosis was developed. In this

paper was shown, how the new mathematical approaches, in particular, neural networks, are

used in medical researches for diagnostics and predictions. The quality of a model

constructed with the help of neural network also was evaluated in this research. We

compared the outcomes obtained by using this model with the outcomes obtained by using

stepwise multiple regression. This comparison obvious shows that the experience of

application of neural networks for prognosis of the first-year child growth has shown a

higher degree of a reliability of obtained inferences.

Abstract. The following describes interactive PC software for creating, tuning and

maintenance of feedforward neural networks. We called this software NNC - Neural Network

Constructor. The system requirements, the users interface and the main algorithms are

described below. NNC v.3.01 is free software available (as soon as short user manual) on

World Wide Web at http://lmgdd.ibmh.msk.su/NNC. Executing NNC software a user can easily

create, tune and exploit neural networks. Several problems have already been successfully

solved using this software. Here we consider one solved problem arising from medicine and

one solved problem arising from biochemistry. In the first case the training set consisted

of 42 observations and described by 11 variables. In the last case the training set

consisted of 150 observations and described by 9 variables. Nevertheless, we have already

solved a lot of problems from the different fields of knowledge using this software.

Contacts information:

Vladimir V. Krepets

Ph.D.,

********@*****.***

Contact this candidate