Deep learning tools for the analysis of movement, identity and behavior


SOCIETY FOR INTEGRATIVE AND COMPARATIVE BIOLOGY
2021 VIRTUAL ANNUAL MEETING (VAM)
January 3 – Febuary 28, 2021

Meeting Abstract


S6-4  Tue Jan 5 11:30 – 12:00  Deep learning tools for the analysis of movement, identity and behavior Mathis, A; EPFL alexander.mathis@epfl.ch

Title: Deep Learning Tools for the Analysis of Movement, Identity & Behavior Alexander Mathis Center for Neuroprosthetics, Brain Mind Institute, School of Life Sciences École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland * alexander.mathis@epfl.ch Quantifying behavior is crucial for many applications across biology. Videography provides easy methods for the observation and recording of animal behavior in diverse settings, yet extracting particular aspects of a behavior for further analysis can be highly time consuming and computationally challenging. I will discuss the latest developments for an efficient method for markerless pose estimation based on transfer learning with deep neural networks that achieves excellent results with minimal training data (Mathis et al., Nature Neuroscience 2018). Furthermore, I will discuss how to directly predict behaviors from video as well as show new architectures for performing re-identification of animals in the lab and the wild. I will illustrate the versatility of these tools for multiple species across a broad collection of behaviors from egg-laying flies, via bears to 3D pose estimation on hunting cheetahs.

the Society for
Integrative &
Comparative
Biology