Invited Talk by Dr. Hari Koduvely on Machine Learning through Bayesian Inference
Bayesian Inference provides a unified framework to incorporate many aspects of real world problems into machine learning models in a natural way. This includes existing domain knowledge about the problem, missing data, latent variables, features and model selection, updating models when new data is observed and so on. However due to the high computational requirements to carry out Bayesian inference, it was not adapted that well for large scale machine learning problems till recently. In this talk I will present basics of Bayesian Inference methods and its application in building machine learning models for recommendation systems using Probabilistic Matrix Factorization. I will also present how Bayesian Probabilistic Matrix Factorization can be scaled to large data sizes using Map-Reduce parallelization.
Dr. Hari Koduvely is a Principal Data Scientist with Samsung R&D Institute, Bangalore, India. His areas of research interests include large scale machine learning, information extraction from text, recommendation systems and Bayesian inference. Prior to Samsung Dr. Hari has worked for Infosys, Amazon and Unilever R&D. He has several publications in the field of data science including a recent book, Learning Bayesian Models with R, published by Packt Publishers UK. Hari has done his PhD in Statistical Physics from Tata Institute of Fundamental Research, Mumbai and post-doctoral research from Weizmann Institute, Israel and Georgia Tech, Atlanta.