Learning multi-class classification problems

Timothy L. H. Watkin , Albrecht Rau , Desiré Bollé , Jort van Mourik

A multi-class perceptron can learn from examples to solve problems whose answer may take several different values. Starting from a general formalism, we consider the learning of rules by a Hebbian algorithm and by a Monte-Carlo algorithm at high temperature. In the benchmark “prototype-problem” we show that a simple rule may be more than an order of magnitude more efficient than the well-known solution, and in the conventional limit is in fact optimal. A multi-class perceptron is significantly more efficient than a more complicated architecture of binary perceptrons.

Domaines: Articles anciens

Timothy L. H. Watkin, Albrecht Rau, Desiré Bollé, Jort van Mourik. Learning multi-class classification problems. Journal de Physique I, 1992, 2 (2), pp.167-180. ⟨10.1051/jp1:1992131⟩. ⟨jpa-00246470⟩