In a statistical decision problem, the use of approximate or estimated priors leads naturally to the question of the continuity of risk in the variable . This continuity condition is explored, and in the finite parameter state case, a characterization in terms of the differentiability of the Bayes envelope is given.
Keywords :
Bayes procedures; Bit rate; Covariance matrix; Entropy; Equations; Linear predictive coding; Matrix decomposition; Reflection; Signal processing; Speech processing; State estimation;