Research

299604_10150269795021876_7084459_nMy academic interests are in cross-modal and cross-domain transfer learning as a precursor or requirement to computational creativity.  In this section, you can find the following:

Interesting Papers (not yet annotated) and my Current Reading Stack.

Updates to my code writing progress towards my dissertation

Premise

It has been shown that it is possible to learn a cross-domain transfer function using a RBM modified to have two visible, input layers and one hidden-layer connected multiplicatively through a third-order tensor.  Training, however, seems messy and extending to a hierarchical or multilayered architecture presents some difficulties.  Training multilayered autoencoders seems straightforward, and recent approaches to training deep learning networks have shown much promise in accuracy and efficiency.

It may be possible to combine the use of the three-way tensors and the clean structure of autoencoders to create deep learning techniques for extending the range between task domains for transfer in reinforcement learning.

 

 

Write a Comment

Comment