A perceptron classifier and corresponding probabilities

Abstract

In this paper a fault tolerant probabilistic kernel version with smoothing parameter of Minsky’s perceptron classifier for more than two classes is sketched. Moreover a probabilistic interpretation of the output is exhibited. The price one has to pay for this improvement appears in the non-determinism of the algorithm. Nevertheless an efficient implementation using for example Java concurrent programming and suitable hardware is shown to be possible. Encouraging preliminary experimental results are presented.