Differentiable Programming

Dominic Danks, Turing PhD Student

Differentiable programming is a programming paradigm in which programs are created so that they can be differentiated, usually via automatic differentiation, allowing the use of gradient based optimisation of parameters in the program. In this lecture, we will examine Google JAX - a high-performance machine learning research library - which provides a framework for easy automatic differentiation of Python/NumPy-based functions.

Topics covered include:

• Principles of differentiable programming
• Google JAX