Composing Graphical Models With Neural Networks with David Duvenaud - TWiML Talk #96

This article summarizes a podcast episode featuring David Duvenaud, discussing his work on combining probabilistic graphical models and deep learning. The focus is on a framework for structured representations and fast inference, with a specific application in automatically segmenting and categorizing mouse behavior from video. The conversation also touches upon the differences between frequentist and Bayesian statistical approaches. The article highlights the practical application of the research and the potential for broader use cases.

• •David Duvenaud discusses a framework that combines probabilistic graphical models and deep learning.
• •The framework is used for structured representations and fast inference.
• •A use case involves automatically segmenting and categorizing mouse behavior from video.