TY - GEN

T1 - Undifferentiated optimization of data sample

AU - Orekhov, Andrey V.

PY - 2017/7/10

Y1 - 2017/7/10

N2 - Obtaining an unbiased data sample is an important task in the statistical analysis of experimental data. The unbiased data sample is a representative data sample. The natural desire is to obtain a representative data sample using computational methods. A procedure for adjusting the structure of the data sample in line with the structure of statistical population is called 'correction of a data sample'. This procedure optimizes data sample, minimizing the difference between a theoretical distribution of control variables and an empirical distribution of control variables. The variables are called control ones if we know the distribution of their spectral values in the statistical population. All of the known methods of adjusting the data sample have significant drawback, as they 'correct' an empirical distribution function, but not the data sample. For example, that refers to IPF algorithm [1], [2]. We discuss an algorithm that corrects sample data rather than their empirical distributions. This algorithm is randomized. An algorithm is called randomized, if the execution of one or several iterations relies on a random rule [3]. The optimization of a data sample carried out with a randomized algorithm cannot be differentiable. This algorithm can be considered as the inhomogeneous Markov chain [4].

AB - Obtaining an unbiased data sample is an important task in the statistical analysis of experimental data. The unbiased data sample is a representative data sample. The natural desire is to obtain a representative data sample using computational methods. A procedure for adjusting the structure of the data sample in line with the structure of statistical population is called 'correction of a data sample'. This procedure optimizes data sample, minimizing the difference between a theoretical distribution of control variables and an empirical distribution of control variables. The variables are called control ones if we know the distribution of their spectral values in the statistical population. All of the known methods of adjusting the data sample have significant drawback, as they 'correct' an empirical distribution function, but not the data sample. For example, that refers to IPF algorithm [1], [2]. We discuss an algorithm that corrects sample data rather than their empirical distributions. This algorithm is randomized. An algorithm is called randomized, if the execution of one or several iterations relies on a random rule [3]. The optimization of a data sample carried out with a randomized algorithm cannot be differentiable. This algorithm can be considered as the inhomogeneous Markov chain [4].

KW - EXPECTED MARGINAL TOTALS

KW - TABLES

UR - http://www.scopus.com/inward/record.url?scp=85027454747&partnerID=8YFLogxK

U2 - 10.1109/CNSA.2017.7973995

DO - 10.1109/CNSA.2017.7973995

M3 - Conference contribution

AN - SCOPUS:85027454747

SP - 229

EP - 231

BT - 2017 Constructive Nonsmooth Analysis and Related Topics (Dedicated to the Memory of V.F. Demyanov), CNSA 2017 - Proceedings

A2 - Polyakova, L. N.

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2017 Constructive Nonsmooth Analysis and Related Topics

Y2 - 22 May 2017 through 27 May 2017

ER -