MATH MODELING

On some properties of randomized machine learning procedures in the presence of noisy data

Yu. S. Popkov

Federal Research Center "Computer Science and Control" of Russian Academy of Sciences, Moscow, Russia

Abstract: We study various models of measuring noises in the procedures of randomized entropy estimation of probability density functions: additive and multiplicative, measuring noises at the input and output of the object’s model. The properties of entropy-optimal probability density functions are studied, it is shown that the measurement noises corresponding to them are heteroscedastic.

Keywords: entropy estimation, density functions, Lagrange multipliers, heteroscedastic noise, variation models.

DOI: 10.14357/20718632230209

Bibliographic databases:

•