Rose: Composable Autodiff for the Interactive Web

Λεπτομέρειες βιβλιογραφικής εγγραφής
Τίτλος: Rose: Composable Autodiff for the Interactive Web
Συγγραφείς: Sam Estep and Wode Ni and Raven Rothkopf and Joshua Sunshine, Estep, Sam, Ni, Wode, Rothkopf, Raven, Sunshine, Joshua
Στοιχεία εκδότη: Schloss Dagstuhl – Leibniz-Zentrum für Informatik 2024
Τύπος εγγράφου: Mixed materials
Περίληψη: 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 work introduces 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.
Όροι ευρετηρίου: Automatic differentiation, differentiable programming, compilers, web, InProceedings, Text, doc-type:ResearchArticle, publishedVersion
DOI: 10.4230.LIPIcs.ECOOP.2024.15
Σύνδεσμος: https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ECOOP.2024.15
Is Part Of LIPIcs, Volume 313, 38th European Conference on Object-Oriented Programming (ECOOP 2024)
Διαθεσιμότητα: Open access content. Open access content
https://creativecommons.org/licenses/by/4.0/legalcode
Σημείωση: application/pdf
English
Other Numbers: DEDAG oai:drops-oai.dagstuhl.de:20864
doi:10.4230/LIPIcs.ECOOP.2024.15
urn:nbn:de:0030-drops-208642
1460281780
Πηγή συνεισφοράς: SCHLOSS DAGSTUHL LEIBNIZ ZENTRUM GMBH
From OAIster®, provided by the OCLC Cooperative.
Αριθμός Καταχώρησης: edsoai.on1460281780
Βάση Δεδομένων: OAIster
Περιγραφή
DOI:10.4230.LIPIcs.ECOOP.2024.15