Automatic Differentiation via Effects and Handlers: An Implementation in Frank
This work presents a novel implementation method for AD, which is incremental as it applies existing effects and handlers concepts to a specific algorithmic family.
The authors tackled the implementation of automatic differentiation (AD) by using effects and handlers, demonstrating this approach in the Frank language and showing how it dynamically creates programs during evaluation.
Automatic differentiation (AD) is an important family of algorithms which enables derivative based optimization. We show that AD can be simply implemented with effects and handlers by doing so in the Frank language. By considering how our implementation behaves in Frank's operational semantics, we show how our code performs the dynamic creation of programs during evaluation.