Dynamic parameter estimation based on minimum cross-entropy method for combining information sources
Abstrakt
When combining information sources, e.g. measuring devices or experts, we deal with two problems: which combining method to choose (linear combination, geometric mean) and how to measure the reliability of the sources, i.e. how to assign the weights to them. Inspired by [Sečkárová, 2013] we introduce a method which overcomes such shortcomings.
Proposed method, based on minimization of the Kullback-Leibler divergence with specific constraints, directly combines data, i.e. probability vectors, thus no additional step to obtain the weights is needed. The detailed description of the proposed method and a comparison with recently introduced dynamic diffusion estimation [Dedecius and Sečkárová, 2013], which heavily depends on the determination of the weights, form the core of this contribution.