Edward is a python library for probabilistic modeling and inference. It is based on tensorflow and leverages the computational graph and tools such as automatic differentiation to automate inference in probabilistic models. This means that users can skip the difficult step of deriving a custom inference algorithm and can use Edward to fit more complex probabilistic models to their data. All they have to do is specify the probabilistic model.
Outline of the talk
First, I will introduce the tensorflow and Edward basics that are necessary to look at a few modeling examples. The examples we will cover include how to fit a Bayesian neural network and an embedding model to real data.
By introducing probabilistic modeling in Edward and giving an overview of how it can be used, I hope to encourage people to use Edward for their data science projects an/or to start contributing to the library.