Reducing the Dimensionality of Data with Neural Networks
Abstract
Highdimensional data can be converted to lowdimensional codes by training a multilayer neural network with a small central layer to reconstruct highdimensional input vectors. Gradient descent can be used for finetuning the weights in such ``autoencoder'' networks, but this works well only if the initial weights are close to a good solution. We describe an effective way of initializing the weights that allows deep autoencoder networks to learn lowdimensional codes that work much better than principal components analysis as a tool to reduce the dimensionality of data.
 Publication:

Science
 Pub Date:
 July 2006
 DOI:
 10.1126/science.1127647
 Bibcode:
 2006Sci...313..504H
 Keywords:

 COMP/MATH