Mackenzie Mathis

Tutorial on “Using deep learning in Neuroscience: Pose estimation for animals”

This tutorial will cover how recent progress in deep learning can be harnessed in the laboratory. Crucially, neuroscientists need tools that can be robust yet work with little training data. Here, I will discuss our progress in developing new toolboxes for animal pose estimation. Specifically, I will discuss how transfer learning enables efficient pose estimation, how we can extract 3D postures to be used in conjunction with neural data, measuring interactions of multiple animals, and how these tools can be used to do real-time analysis. We will also get hands-on experience using our toolbox – DeepLabCut – in this tutorial to explore how it works, but crucially, to discuss what is required to make tools accessible to neuroscientists who might not be experts in machine learning.

Suggested reading:

Protocol:

Participants can use our demo dataset and install nothing, or, if you wish to analyze your own video(s), you need to install anaconda on your laptop (https://www.anaconda.com/distribution/) and download the appropriate CPU-based “conda” file from our website: www.deeplabcut.org

(please email or see Mackenzie Mathis if you have questions on this step!)