Large-scale Latent Variable Modeling: Inference Algorithms and Applications
===========================================================================

Latent variable models have recently gained lots of attentions due to
their flexibility of modeling complex data. These models are widely
used for latent theme discovery, browsing and recommendations. In this
talk, I will describe some of my recent research on large-scale
inference algorithms and novel applications for latent variable
models.

First I will describe how to fit large-scale latent variable models
(like topic models) with millions of documents on a single machine
using stochastic variational inference. Stochastic variational
inference uses the idea of stochastic optimization and in the
meantime, carefully considers the curvature induced by the variational
distribution. This results in a simple yet efficient algorithm by only
visiting a small batch of data samples at each update.

Second, I will describe collaborative topic models for article
recommendations. Unlike traditional matrix factorization, these models
give interpretable dimensions to user interests and article profiles.
It also naturally solves the cold-start problem and can form
recommendations about sparsely rated or previously unrated articles. I
will also discuss how we can embed this idea into many other
applications to achieve better predictive performance but still
maintain the interpretability of the model.

Joint work with David Blei, Matt Hoffman and John Paisley.


---Bio---
Chong Wang is a project scientist in Machine Learning Department,
Carnegie Mellon University. He received his PhD from Princeton
University in 2012, advised by David M. Blei. He research has won
several awards, including a best student paper award at KDD 2011, a
notable paper award at AISTATS 2011 and a best student paper award
honorable mention at NIPS 2009. He received the Google Ph.D.
Fellowship for machine learning in 2010. He is also a Siebel Scholar,
class of 2012. His thesis was nominated for ACM Doctoral Dissertation
Award by Princeton University in 2012.