Deriving forward and reverse mode AD
We will derive forward and reverse mode automatic differentiation (AD) for pure, straight-line programs by example.
I am a postdoctoral researcher with the Effect Handler Oriented Programming project at the University of Edinburgh, which is led by Sam Lindley.
My interests include programming languages, categorical semantics, effects and handlers, differentiable programming, and probabilistic programming. I have recently finished my PhD, supervised by Chris Heunen, with my thesis entitled Automatic Differentiation via Effects and Handlers. In my thesis I provide concrete implementations of automatic differentiation algorithms in various effect handler languages and prove correctness for some using enriched category theory and denotational semantics.
We will derive forward and reverse mode automatic differentiation (AD) for pure, straight-line programs by example.
In this post we will show that the local state monad of (Plotkin & Power, 2002) (which was suggested by Peter O’Hearn) has rank.