Alternating minimization and Boltzmann machine learning

W. Byrne
1992 IEEE Transactions on Neural Networks  
Paining a Boltzmann machine with hidden units is appropriately treated in information geometry using the information divergence and the technique of alternating minimization. The resulting algorithm is shown to be closely related to gradient descent Boltzmann machine learning rules, and the close relationship of both to the EM algorithm is described. An iterative proportional fitting procedure for training machines without hidden units is described and incorporated into the alternating minimization algorithm.
doi:10.1109/72.143375 pmid:18276461 fatcat:c6xk34xyjfaffdr46qysaw7d4y