Academic Journal
Rose: Composable Autodiff for the Interactive Web (Artifact)
| Title: | Rose: Composable Autodiff for the Interactive Web (Artifact) |
|---|---|
| Authors: | Estep, Sam, Ni, Wode, Rothkopf, Raven, Sunshine, Joshua |
| Contributors: | Sam Estep and Wode Ni and Raven Rothkopf and Joshua Sunshine |
| Publisher Information: | Schloss Dagstuhl – Leibniz-Zentrum für Informatik |
| Publication Year: | 2024 |
| Collection: | DROPS - Dagstuhl Research Online Publication Server (Schloss Dagstuhl - Leibniz Center for Informatics ) |
| Subject Terms: | Automatic differentiation, differentiable programming, compilers, web |
| Description: | Reverse-mode automatic differentiation (autodiff) has been popularized by deep learning, but its ability to compute gradients is also valuable for interactive use cases such as bidirectional computer-aided design, embedded physics simulations, visualizing causal inference, and more. Unfortunately, the web is ill-served by existing autodiff frameworks, which use autodiff strategies that perform poorly on dynamic scalar programs, and pull in heavy dependencies that would result in unacceptable webpage sizes. This document accompanies the research paper introducing Rose, a lightweight autodiff framework for the web using a new hybrid approach to reverse-mode autodiff, blending conventional tracing and transformation techniques in a way that uses the host language for metaprogramming while also allowing the programmer to explicitly define reusable functions that comprise a larger differentiable computation. We demonstrate the value of the Rose design by porting two differentiable physics simulations, and evaluate its performance on an optimization-based diagramming application, showing Rose outperforming the state-of-the-art in web-based autodiff by multiple orders of magnitude. |
| Document Type: | article in journal/newspaper |
| File Description: | application/pdf |
| Language: | English |
| Relation: | Is Part Of DARTS, Volume 10, Issue 2, Special Issue of the 38th European Conference on Object-Oriented Programming (ECOOP 2024); https://drops.dagstuhl.de/entities/document/10.4230/DARTS.10.2.7 |
| DOI: | 10.4230/DARTS.10.2.7 |
| Availability: | https://doi.org/10.4230/DARTS.10.2.7 https://nbn-resolving.org/urn:nbn:de:0030-drops-209053 https://drops.dagstuhl.de/entities/document/10.4230/DARTS.10.2.7 |
| Rights: | https://creativecommons.org/licenses/by/4.0/legalcode |
| Accession Number: | edsbas.6619797C |
| Database: | BASE |
| DOI: | 10.4230/DARTS.10.2.7 |
|---|