*Result*: Minimization of Systematic and Random Errors in Linear Estimation Problems for Big Data Processing.
*Further Information*
*The problem of processing distributed results from multiple experiments while flexibly controlling both random and systematic estimation errors is studied. The approach is based on transforming the results of individual experiments into a special "canonical" form, then combining the resulting fragments of canonical information and constructing an optimal family of estimates based solely on the accumulated canonical information. A detailed study of the optimal family of estimates is also conducted. The use of canonical information significantly facilitates the study and allows for a detailed characterization of the error structure in operator form. The key properties of canonical information and the advantages of its use within the MapReduce distributed computing model for big data are discussed. [ABSTRACT FROM AUTHOR]*