ENTROPY MAXIMIZATION NETWORKS - AN APPLICATION TO BREAST-CANCER PROGNOSIS

This paper describes two artificial neural network architectures for c onstructing maximum entropy models using multinomial distributions, Th e architectures presented maximize entropy in two ways: by the use of the partition function (which involves the solution of simultaneous po lynomial equations) and by constrained gradient ascent, Results compar ing the convergence properties of these two architectures are presente d. The practical use of these two architectures as a method of inferen ce is illustrated by an application to the prediction of metastases in early breast cancer patients, To assess the predictive accuracy of th e maximum entropy models, we compared the results with those obtained by the use of the multilayer perceptron and the probabilistic neural n etwork.