Powered by Indico
v3.2.8
Flatiron Institute community members are cordially invited to a CCN Seminar with Matt Whiteway, Associate Research Scientist at Zuckerman Instituteat Columbia University and Data Scientist at the International Brain Lab His talk title and abstract are below.
To schedule a meeting with Matt during his visit, please click here or be in touch with Jessica Hauser at jhauser@flatironinstitute.org.
Title:Partitioning variability in animal behavioral videos using semi-supervised variational autoencoders
Abstract:
The study of animal behavior has been transformed over the past decade by advances in computer vision algorithms. Most of these algorithms are fully supervised (e.g. pose estimation) or fully unsupervised (e.g. PCA, autoencoders). I will describe an autoencoder-based algorithm that combines the strengths of each approach to produce interpretable, low-dimensional representations of behavioral videos. I'll then demonstrate how these interpretable representations can facilitate downstream behavioral and neural analyses on several datasets. Importantly, I will also discuss some of the practical limitations of this approach that arise from the presence of nuisance variables like camera angle and contrast levels, and how these issues have been addressed by myself and others through modifications to the model structure and loss functions.
Matthew Turner