This shows you the differences between two versions of the page.

Link to this comparison view

correlational_network [2016/12/13 19:09] (current)
Line 1: Line 1:
 +Correlational Neural Networks
 +Common Representation Learning (CRL), wherein different descriptions (or
 +views) of the data are embedded in a common subspace, is one way of achieving
 +Transfer Learning. Two popular paradigms here are Canonical Correlation Analysis
 +(CCA) based approaches and Autoencoder (AE) based approaches. Each of
 +these approaches has its own advantages and disadvantages. For example, while
 +CCA based approaches outperform AE based approaches for the task of transfer
 +learning, they are not as scalable as the latter. In this work we propose an
 +AE based approach called Correlational Neural Network (CorrNet), that explicitly
 +maximizes correlation among the views when projected to the common subspace.
 +Through experiments,​ we demonstrate that the proposed CorrNet is better than the
 +above mentioned approaches with respect to its ability to learn correlated common
 +representations that are useful for Transfer Learning.