DMCA
Deep Canonical Correlation Analysis
Cached
Download Links
| Citations: | 20 - 4 self |
BibTeX
@MISC{Andrew_deepcanonical,
author = {Galen Andrew and Jeff Bilmes and Karen Livescu},
title = {Deep Canonical Correlation Analysis},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
We introduce Deep Canonical Correlation Analysis (DCCA), a method to learn complex nonlinear transformations of two views of data such that the resulting representations are highly linearly correlated. Parameters of both transformations are jointly learned to maximize the (regularized) total correlation. It can be viewed as a nonlinear extension of the linear method canonical correlation analysis (CCA). It is an alternative to the nonparametric method kernel canonical correlation analysis (KCCA) for learning correlated nonlinear transformations. Unlike KCCA, DCCA does not require an inner product, and has the advantages of a parametric method: training time scales well with data size and the training data need not be referenced when computing the representations of unseen instances. In experiments on two real-world datasets, we find that DCCA learns representations with significantly higher correlation than those learned by CCA and KCCA. We also introduce a novel non-saturating sigmoid function based on the cube root that may be useful more generally in feedforward neural networks.
Keyphrases
deep canonical correlation analysis real-world datasets training data training time unseen instance resulting representation nonlinear transformation data size feedforward neural network complex nonlinear transformation inner product novel non-saturating sigmoid function cube root parametric method total correlation nonlinear extension linear method canonical correlation analysis
