Application of expectation consistent approximate inference
published: Feb. 25, 2007, recorded: January 2005, views: 3138
Report a problem or upload filesIf you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.
I will discuss two types of applications of an approximate inference technique (EC = expectation consistent) recently developed together with Ole Winther. The EC method is an extension of the TAP (Thouless, Anderson & Palmer) approach which originated in the field of disordered materials and which has been further developed to become applicable to a variety of scenarios in probabilistic modelling & machine laerning. My first application (joint work with Doerthe Malzahn) deals with an approximation to resampling methods (such as the bootstrap) which allows to estimate eg generalization errors in supervised learning. While the exact resampling approach requires the drawing of many samples from the training data and a costly repeated retraining of the model, the approximation attempts an analytic average which combines the replica trick and an inference method which can be performed much faster. In the second application (ongoing work) I discuss the scenario of many solutions to the EC framework and the possibility of averaging them using Parisi's hierarchical scheme.
Link this pageWould you like to put a link to this lecture on your homepage?
Go ahead! Copy the HTML snippet !