This Major Qualifying Project dives into the classic problem in machine learning where we want to predict one random variable from another random variable and. Having taken inspiration from Canonical Correlation Analysis (linear methods) and autoencoders (non-linear methods), we propose various algorithms to improve the predictability between two random variables. Our work provides interesting insights into how to train neural networks to reconstruct one variable from another. In particular, it underlines the importance of obtaining an information-rich latent representation of each variable. Moreover, we demonstrated the importance of the hidden representation of an autoencoder and the benefits of utilizing random data correctly towards our problem.

• This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator

• Pham, Minh N.

Publisher

• Worcester Polytechnic Institute

Identifier

• 23156
• E-project-050621-152256

Advisor

• Paffenroth, Randy Clinton
• Whitehill, Jacob Richard

Year

• 2021

Date created

• 2021-05-06

Resource type

• Major Qualifying Project

Major

• Computer Science

Rights statement

• In Copyright

Last modified

• 2021-08-29