Key points are not available for this paper at this time.
Automatic differentiation (AD) has become the backbone for a new wave of optimization-driven domains such as computer graphics and machine learning over the past decade. However, existing AD systems face limitations, either lacking support for in-browser development or failing to harness more recent, compilerbased approaches to achieve both expressiveness and size-preserving differentiation. This work introduces Rose, a portable, extensible AD library that runs on the web. Rose allows users to write opaque functions with custom derivatives and supports dynamic construction of AD functions. We integrated Rose into two differentiable simulations and a diagram authoring tool to demonstrate the utility of Rose's design. Finally, we show that Rose is 173x as fast as TensorFlow.js in compiling and running a benchmark suite of optimized diagrams.
Estep et al. (Tue,) studied this question.