SEMINAR:Probabilistic Programming and Bayesian Inference in Biomedicine

SEMINAR:Probabilistic Programming and Bayesian Inference in Biomedicine

LISTEN

Speaker: Aybars Can Acar

Title:  Probabilistic Programming and Bayesian Inference in Biomedicine

Date/Time: 25 November 2020/ 13:40 - 14:30

Zoom: Meeting ID 910 3770 2809

Passcode:  gradsem

Abstract:Probabilistic programming is a relatively new approach to problem solving where we define arbitrarily large and sophisticated probabilistic models as generative programs. We can then infer latent variables of interest by sampling from the outputs of these programs, allowing us to solve complex inference problems without having to `invert` them first. This, combined with Bayesian inference, results in a powerful way of reasoning about -- and testing hypotheses on -- poorly understood phenomena, especially under uncertainty.In this talk, I will give an overview of probabilistic programming and its evolution over the last five years, the technical difficulties involved (such as time costs of sampling) and possible remedies (e.g. variational methods and parallelization). I will also discuss the applications of this methodology to biomedical problems, with specific examples on gene expression analysis, prediction of aortic aneurysm growth, and COVID-19 epidemic projection.

Bio:Aybar C. Acar is a faculty member at the Middle East Technical University (METU) Graduate School of Informatics. He received his BS and MS degrees in Chemical Engineering from METU, and his PhD in Computer Science from George Mason University. His primary areas of interest are probabilistic modeling, machine learning, transcriptomics and systems biology. He is currently the co-director of the Cancer Systems Biology Laboratory at METU