University of Cambridge > > Machine Learning @ CUED > Moment matching for latent variable models: from ICA to LDA and CCA

Moment matching for latent variable models: from ICA to LDA and CCA

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Zoubin Ghahramani.

Moment matching is a traditional alternative to maximum likelihood for parameter estimation in probabilistic models. For certain latent variable models, this has recently led to parameter estimation algorithms with theoretical guarantees. While independent component analysis (ICA) was the first semi-parametric model to be considered twenty years ago, this has been recently extended to latent Dirichlet Allocation (LDA), which is a parametric model for discrete data. In this talk I will present (a) a semi-parametric extension of LDA which, beyond making fewer modelling assumptions, leads to simpler estimation through moment matching techniques, and (b) an extension to multi-view models such as canonical correlation analysis (CCA). (Joint work with Anastasia Podosinnikova and Simon Lacoste-Julien).

This talk is part of the Machine Learning @ CUED series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2024, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity