Statistical Mechanics of Unsupervised Learning
Michael L. Biehl, A Mietzner · Europhysics Letters (EPL) · 1993
We study two different unsupervised learning strategies for a single-layer perceptron. The environment provides a set of unclassified training examples, which belong to two different classes, depending on their overlap with an N -dimensional concept vector. By means of a statistical-mechanics analysis, using the replica method, we investigate how well the perceptron infers the unknown structure from the input data.