A Probabilistic Algorithm for Calculating Similarities

Resumo

In this paper, we describe a new probabilistic algorithm for calculating hypotheses as the results of similarities between training examples for a machine learning problem based on a binary similarity operation. Unlike previously proposed probabilistic algorithms, the order of accounting for training examples is fixed for all hypotheses. This algorithm is useful for implementation using a GPGPU. The main result of this paper is the independence of the order of the appearance of training examples of the probabilities of each similarity in the sample.