MIA: Martin Jankowiak, Bayesian methods for adaptive experimental design

Models, Inference and Algorithms
Broad Institute of MIT and Harvard
February 10, 2021

Bayesian methods for adaptive experimental design

Martin Jankowiak
Pyro Team, Broad Institute

As expensive, high-throughput experiments become routine, it is increasingly important to make efficient use of limited experimental resources. Unfortunately, in complex settings human intuition may not be up to the task of making suitable choices for the many design parameters that enter into intricate experiments. Bayesian optimal experimental design (OED) is a principled information-theoretic framework for automating certain aspects of experimental design. What makes OED particularly attractive is that it can enable adaptive experiments in which data from previous rounds informs the experimental design used in subsequent rounds. We give an introduction to the principles that underlie OED and show how recent advances in black-box variational inference make OED suitable for practical use.

For more information on the Broad Institute and Models, Inference and Algorithms visit: https://www.broadinstitute.org/mia

Copyright Broad Institute, 2021. All rights reserved.

Видео MIA: Martin Jankowiak, Bayesian methods for adaptive experimental design канала Broad Institute