An Exactly Solvable Model of Unsupervised Learning

@inproceedings{InstituteBlegdamsvejAnES,
title={An Exactly Solvable Model of Unsupervised Learning},
author={Niels Bohr InstituteBlegdamsvej}
}

Niels Bohr InstituteBlegdamsvej

A model for unsupervised learning from N {dimensional data is studied. Random training examples are drawn such that the distribution of their overlaps with a vector B 2 IR N is a mixture of two Gaussians of unit width and a separation. A student vector is generated by an online algorithm, using each example only once. The evolution of its overlap with B can be calculated exactly in the thermodynamic limit N ! 1. As a speciic example, a learning algorithm closely related to Oja's rule is… CONTINUE READING