Automatic Bayesian Learning Methods
From Anita Borg Institute Wiki
Date/Time: October 19, 2007, 10-10:20am
Towards Automatic Bayesian Learning Methods
Presenter:
Jo-Anne Ting, University of Southern California, Los Angeles, CA
Summary:
We introduce a toolbox of automatic methods that allow us to perform high-dimensional linear regression with both noise-contaminated inputs and outputs, detect outliers in observed data in real-time (i.e., in data streams), and learn the optimal values of spatial distance metrics of local models. The use of Bayesian statistics and variational approximation techniques allows us to perform automatic regularization and achieve fast inference. These methods do not use any cross-validation, sampling or experimental tuning to learn the optimal values of open parameters. Instead, they learn the optimal values of open parameters automatically.
For more information, please refer to the following links: