Title : 
Alternating minimization and Boltzmann machine learning
         
        
        
            Author_Institution : 
Dept. of Electr. Eng., Maryland Univ., College Park, MD, USA
         
        
        
        
        
            fDate : 
7/1/1992 12:00:00 AM
         
        
        
        
            Abstract : 
Training 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
         
        
            Keywords : 
iterative methods; learning systems; minimisation; neural nets; Boltzmann machine learning; alternating minimization; hidden units; information divergence; information geometry; iterative proportional fitting; learning rules; neural nets; Information geometry; Iterative algorithms; Machine learning; Machine learning algorithms; Minimization methods; Neural networks; Particle measurements; Stochastic processes; Symmetric matrices; Temperature;
         
        
        
            Journal_Title : 
Neural Networks, IEEE Transactions on