Bayesian Sparse Principal Component Analysis

Several real-life high dimension datasets can be reasonably represented as a linear combination of a few sparse vectors. Succinct representation of such data with a few selected variables is highly desirable for such cases. A Bayesian setup is useful because the limitation of knowing a limited number of  high dimensional data points can be alleviated by well-designed domain-specific priors. WNCG Prof. Joydeep Ghosh, his student Rajiv Khanna, and WNCG alumnus Oluwasanmi Koyejo, currently at Stanford, are developing scalable Bayesian PCA models to extract sparse components from large datasets using a novel constrained inference framework. Results obtained so far show clear superiority as compared to a large list of standard baselines.

This work will be presented at AISTATS 2015.

The preprint is available for view by WNCG Industral Affiliates only.