Applying Measures of Dependence Based on the Symmetric Rényi Divergence to the Statistical Linearization. Issue 1 (July 2017)

• Record Type: Journal Article
• Authors: Chernyshov, K.R.
• Is Part Of: IFAC-PapersOnLine. Volume 50:Issue 1(2017)
• Page Start: 8347
• Page End: 8354
• Publication Date: 2017-07
• Subjects: Statistical linearization -- Information theory -- Nonlinear systems -- Rényi entropy -- System identification -- Divergence measures -- Consistent measures of dependence -- Global optimization -- Parameter estimation -- Criterion functions; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2017.08.1557
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: The paper presents an approach to the statistical linearization of input/output mappings of stochastic systems, with emphasizing the significance of applying consistent (in accordance to the A.N. Kolmogorov terminology) measures of dependence of random values. Within such a problem statement, the criterion function is constructed by applying the Rényi symmetric divergence leading to the corresponding consistent measure of dependence, the Rényi symmetric mutual information, which combines both properties of the Kullback-Leibler divergence (expansion on the sum of corresponding summands) and the Rényi entropy (logarithm of integral instead of integral of logarithm), what is of particular importance, from the system identification point of view, under constructing sample estimates, especially those of based on observation of dependent data.