This work reports the results obtained with the application of High Or
der Boltzmann Machines without hidden units to construct classifiers f
or some problems that represent different learning paradigms. The Bolt
zmann Machine weight updating algorithm remains the same even when som
e of the units can take values in a discrete set or in a continuous in
terval. The absence of hidden units and the restriction to classificat
ion problems allows for the estimation of the connection statistics, w
ithout the computational cost involved in the application of simulated
annealing. In this setting, the learning process can be sped up sever
al orders of magnitude with no appreciable loss of quality of the resu
lts obtained.