Online Bayesian Inference for Latent Ability Models

Latent ability models relate a set of observed variables to a set of latent ability variables. It includes the paired and multiple comparison models, the item response theory (IRT) models, etc. In this talk, first I will present an online Bayesian approximate method for online gaming analysis with paired and multiple comparison models. Experiments on game data show that accuracy of our online algorithm is competitive with a state of the art systems such as TrueSkill, but the running time as well as the code as much shorter. We also compare our method with Glicko rating system, designed for rating chess players. Next, I will give an extension of this Bayesian approach to IRT models with application to Internet ratings data. The proposed method is based on a variant of Stein's identity. I will also briefly describe how this identity can be applied to derive the Kalman gain.