Beyond Deep Learning – Differentiable Programming with Flux

Abstract: Deep learning is a rapidly evolving field, and models are increasingly complex. Recently, researchers have begun to explore ""differentiable programming"", a powerful way to combine neural networks with traditional programming. Differentiable programs may include control flow, functions and data structures, and can even incorporate ray tracers, simulations and scientific models, giving us even unprecedented power to find subtle patterns in our data.

This workshop will show you how this technique, and particularly Flux – a state-of-the-art deep learning library – is impacting the machine learning world. We will show you how Flux makes it easy to create traditional deep learning models, and explain how the flexibility of the Julia language allows complex physical models can be optimised by the same architecture. We'll outline important recent work and show how Flux allows us to easily combine neural networks with tools like differential equations solvers.

Bio: Jeff is one of the creators of Julia, co-founding the project at MIT in 2009 and eventually receiving a Ph.D. related to the language in 2015. He continues to work on the compiler and system internals, while also working to expand Julia’s commercial reach as a co-founder of Julia Computing, Inc.