Lately, Metis Sr. Data Scientist Chad Scherrer has been exploring Pyro, a recent development in probabilistic programming from Uber AI Labs. In a blog post on the topic, he wrote that Pyro is an "exciting development that has a huge potential for large-scale applications."
But before diving into those, he uses his blog to take a step back, particularly with beginners in mind.
"In any technical writing, it's common (at least for me) to realize I need to add some introductory material before moving on. In writing about Pyro, this happened quite a bit, to the point that it warranted this post as a kind of warm-up," writes Scherrer.
He begins by noting that Pyro bills itself as "Deep Universal Probabilistic Programming," but acknowledges that not everyone will know what that really means. So he deconstructs it, starting with Deep, moving on to Probabilistic Programming, and then tackling Universal before diving into Bayesian and Variational Inference.
"While deep learning has generated a lot of mainstream excitement, probabilistic programming is still dramatically underused, and universal probabilistic programming even more so. Pyro's combination of these with scalable and flexible variational inference has the potential to change that. It's notoriously difficult to predict the influence of a new software library, but Pyro is certainly one to keep an eye on," he writes.
Dig into the full blog post here.
