Maximum Partial Likelihood Estimation with Perceptrons

We show the equivalence of two techniques of time series modeling/prediction; (ii) perceptron learning of probability distribution of the truth value of a proposition from first order stochastic density approximations, (ii) Maximum Partial Likelihood (MPL) estimation of the parameters of a logistic regressive model for binary time series. This result provides large training set characteristics for the approximate Kullback-Leibler relative entropy learning scheme.