Poster # M19

The Infinite Hierarchical Factor Regression Model
Piyush Rai & Hal Daumé III, University of Utah Factor Analysis: Explaining Data, Factor Regression: Using factors for making prediction Our Model: Addresses 3 fundamental shortcomings of standard factor analysis

Fixed numberof factors (K) Independent factors All features assumed relevant for FA


K assumed unbounded (via Indian Buffet Process - IBP) Factor constrained by hierarchy (via Kingman's Coalescent) Feature selection (via row sparsity in IBP)

+

Predictive model (based on factors)

A Nonparametric Bayesian Model Experimental results: Gene-Pathway Modeling, Prediction from gene-expression data



Come, visit our poster if you care about: nonparametric Bayesian, computational biology