Asymptotic Bayesian Generalization Error in Topic Model and Stochastic Matrix Factorization

Abstract
We analyze the asymptotic behavior of the Bayesian generalization error in the topic model. Through a theoretical analysis of the maximum pole of the zeta function (real log canonical threshold) of the topic model, we obtain an upper bound of the Bayesian generalization error and the free energy in the topic model and stochastic matrix factorization (SMF; it can be regarded as a restriction of the non-negative matrix factorization). We show that the generalization error in the topic model and SMF becomes smaller than that of regular statistical models if Bayesian inference is attained.
View on arXivComments on this paper