Probabilistic programming: A new paradigm in machine learning – Københavns Universitet

Probabilistic programming: A new paradigm in machine learning

COPLAS TALK by Thomas Hamelryck BIO and DIKU, University of Copenhagen

Date/time: May 18th, 2018, 2:01-3:00 p.m.
Location: Universitetsparken 5, Room 01-0-029

Abstract

Probabilistic Programming is the next big thing in machine learning, next to Deep Learning and Big Data. By combining general purpose programming with probabilistic modelling, probabilistic programming makes it possible to perform complex statistical reasoning with minimum programming efforts and high computational efficiency. The main idea is (a) to formulate a computer program, often using a standard programming language such as Python, that (b) models a specific problem of inference (medical, financial and so on) and (c) perform statistical inference simply by executing the program. This has been made possible by recent breakthroughs in automated inference (including automated advanced sampling methods and variational inference) and numerical computing (including large software frameworks such as Theano and Tensorflow). Probabilistic programming makes it even possible to implement Deep Learning methods that are based on Bayesian principles that deliver reliable estimates of uncertainty. By decoupling statistical modelling (i.e. formulating a question) from statistical inference (i.e. answering the question), probabilistic programming is expected to dramatically extend the scope of probabilistic reasoning in technology, science, finance, society and medicine.

Bio

Thomas Hamelryck has an MSc in engineering, and a PhD in macromolecular crystallography from the Free University of Brussels (VUB). After his PhD, he moved into the area of Bayesian machine learning, which he mostly applied to bioinformatics and the protein folding problem. He is the main editor of the only textbook on Bayesian modelling of protein structure (Springer, 2012). Currently, his main interest lies in the application of Bayesian probabilistic programming and deep learning methods to problems in science, technology and medicin.

HostFritz Henglein (DIKU, tel. +45-30589576

All are welcome. No registration required.

COPLAS stands for: Copenhagen Programming Language Seminar