The Virtual Brain

Eager to know more on Probabilistic modelling, identifiability and degeneracy, Bayesian workflows and Hierarchical modelling?

Register for a 3-day workshop organised by the Institute de Neurosciences des Systemes on "Bayesian Data Analysis" with Michael Betancourt!

About this workshop

Due to the importance and common use of Bayesian framework for inference and prediction, the advanced techniques in probabilistic programming languages to overcome the inference difficulties with big data complemented with big models have been receiving increasing attention in this context.

Stan is a popular platform for facilitating inference, providing an expressive modelling language and implementing state-of-the-art algorithms to draw subsequent Bayesian inferences.

We are very happy to announce the INS courses on "Bayesian Data Analysis", with Michael Betancourt, a core developer of Stan and expert on Hamilton Monte Carlo. The courses begin by surveying Bayesian inference, Bayesian computation and a principled introduction to Stan.

With a solid foundation, we will move onto to the elements of a robust Bayesian workflow in practice and then continue to the problem of interest to neuroscientist such as source localization in Neuroimaging.

The courses are highly interactive, with exercises demonstrating a principled Bayesian workflow and range of modelling techniques run in Python environment.

Courses run for 3 days and include material spanning Probabilistic modelling, identifiability and degeneracy, Bayesian workflows and Hierarchical modelling.

Prerequisites

The course will assume familiarity with the basics of calculus, linear algebra, and probability theory.

For a self-contained introduction to the latter please review my probability theory, conditional probability theory, and common probability densities case studies. The last will be particularly relevant.

Day 1, May 17th

9:30 - 12:30
Lecture on probabilistic modelling

• Modelling and inference
• Generative Modeling

11:00 - 11:30
Coffee break

13:30 - 17:00
Exercises on modelling, Poisson model of spike counting

Day 2, May 18th

9:30 - 12:30
Identifiability and degeneracy lecture

• Robust workflows for Bayesian modelling
• Identifiability and degeneracy

End of morning
examples and discussion

11:00 - 11:30
Coffee break

13:30 - 17:00
Model building Bayesian workflow lecture

• Bayesian model building workflow
• Hands on modelling: MEG/sEEG source analysis

15:00 - 17:00
Exercises and discussion

The nominal material is based around the Poisson progression in the case study "Towards A Principled Bayesian Workflow"

Day 3, May 19th

9:30 - 12:30
Hierarchical modeling lecture and exercises from the first 2 days

End of morning
examples and discussion

11:00 - 11:30
Coffee break

13:30 - 17:00
Exercises on modelling: temporal models with ODEs

Open discussion with participant models, datasets and questions