DROPS

Document

DOI: 10.4230/LIPIcs.ECOOP.2024.15

Rose: Composable Autodiff for the Interactive Web

Authors: Sam Estep, Wode Ni, Raven Rothkopf, and Joshua Sunshine

Published in: LIPIcs, Volume 313, 38th European Conference on Object-Oriented Programming (ECOOP 2024)

Abstract

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.

Cite as

Sam Estep, Wode Ni, Raven Rothkopf, and Joshua Sunshine. Rose: Composable Autodiff for the Interactive Web. In 38th European Conference on Object-Oriented Programming (ECOOP 2024). Leibniz International Proceedings in Informatics (LIPIcs), Volume 313, pp. 15:1-15:27, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2024)

Copy BibTex To Clipboard

@InProceedings{estep_et_al:LIPIcs.ECOOP.2024.15,
  author =	{Estep, Sam and Ni, Wode and Rothkopf, Raven and Sunshine, Joshua},
  title =	{{Rose: Composable Autodiff for the Interactive Web}},
  booktitle =	{38th European Conference on Object-Oriented Programming (ECOOP 2024)},
  pages =	{15:1--15:27},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-341-6},
  ISSN =	{1868-8969},
  year =	{2024},
  volume =	{313},
  editor =	{Aldrich, Jonathan and Salvaneschi, Guido},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ECOOP.2024.15},
  URN =		{urn:nbn:de:0030-drops-208642},
  doi =		{10.4230/LIPIcs.ECOOP.2024.15},
  annote =	{Keywords: Automatic differentiation, differentiable programming, compilers, web}
}

Document

Artifact

DOI: 10.4230/DARTS.10.2.7

Rose: Composable Autodiff for the Interactive Web (Artifact)

Authors: Sam Estep, Wode Ni, Raven Rothkopf, and Joshua Sunshine

Published in: DARTS, Volume 10, Issue 2, Special Issue of the 38th European Conference on Object-Oriented Programming (ECOOP 2024)

Abstract

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.

Cite as

Sam Estep, Wode Ni, Raven Rothkopf, and Joshua Sunshine. Rose: Composable Autodiff for the Interactive Web (Artifact). In Special Issue of the 38th European Conference on Object-Oriented Programming (ECOOP 2024). Dagstuhl Artifacts Series (DARTS), Volume 10, Issue 2, pp. 7:1-7:4, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2024)

Copy BibTex To Clipboard

@Article{estep_et_al:DARTS.10.2.7,
  author =	{Estep, Sam and Ni, Wode and Rothkopf, Raven and Sunshine, Joshua},
  title =	{{Rose: Composable Autodiff for the Interactive Web (Artifact)}},
  pages =	{7:1--7:4},
  journal =	{Dagstuhl Artifacts Series},
  ISBN =	{978-3-95977-342-3},
  ISSN =	{2509-8195},
  year =	{2024},
  volume =	{10},
  number =	{2},
  editor =	{Estep, Sam and Ni, Wode and Rothkopf, Raven and Sunshine, Joshua},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/DARTS.10.2.7},
  URN =		{urn:nbn:de:0030-drops-209053},
  doi =		{10.4230/DARTS.10.2.7},
  annote =	{Keywords: Automatic differentiation, differentiable programming, compilers, web}
}

Document

DOI: 10.4230/LIPIcs.ITP.2023.24

An Extensible User Interface for Lean 4

Authors: Wojciech Nawrocki, Edward W. Ayers, and Gabriel Ebner

Published in: LIPIcs, Volume 268, 14th International Conference on Interactive Theorem Proving (ITP 2023)

Abstract

Contemporary proof assistants rely on complex automation and process libraries with millions of lines of code. At these scales, understanding the emergent interactions between components can be a serious challenge. One way of managing complexity, long established in informal practice, is through varying external representations. For instance, algebraic notation facilitates term-based reasoning whereas geometric diagrams invoke spatial intuition. Objects viewed one way become much simpler than when viewed differently. In contrast, modern general-purpose ITP systems usually only support limited, textual representations. Treating this as a problem of human-computer interaction, we aim to demonstrate that presentations - UI elements that store references to the objects they are displaying - are a fruitful way of thinking about ITP interface design. They allow us to make headway on two fronts - introspection of prover internals and support for diagrammatic reasoning. To this end we have built an extensible user interface for the Lean 4 prover with an associated ProofWidgets 4 library of presentation-based UI components. We demonstrate the system with several examples including type information popups, structured traces, contextual suggestions, a display for algebraic reasoning, and visualizations of red-black trees. Our interface is already part of the core Lean distribution.

Cite as

Wojciech Nawrocki, Edward W. Ayers, and Gabriel Ebner. An Extensible User Interface for Lean 4. In 14th International Conference on Interactive Theorem Proving (ITP 2023). Leibniz International Proceedings in Informatics (LIPIcs), Volume 268, pp. 24:1-24:20, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2023)

Copy BibTex To Clipboard

@InProceedings{nawrocki_et_al:LIPIcs.ITP.2023.24,
  author =	{Nawrocki, Wojciech and Ayers, Edward W. and Ebner, Gabriel},
  title =	{{An Extensible User Interface for Lean 4}},
  booktitle =	{14th International Conference on Interactive Theorem Proving (ITP 2023)},
  pages =	{24:1--24:20},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-284-6},
  ISSN =	{1868-8969},
  year =	{2023},
  volume =	{268},
  editor =	{Naumowicz, Adam and Thiemann, Ren\'{e}},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ITP.2023.24},
  URN =		{urn:nbn:de:0030-drops-183991},
  doi =		{10.4230/LIPIcs.ITP.2023.24},
  annote =	{Keywords: user interfaces, human-computer interaction, Lean}
}

Document

DOI: 10.4230/LIPIcs.ECOOP.2021.3

Gradual Program Analysis for Null Pointers

Authors: Sam Estep, Jenna Wise, Jonathan Aldrich, Éric Tanter, Johannes Bader, and Joshua Sunshine

Published in: LIPIcs, Volume 194, 35th European Conference on Object-Oriented Programming (ECOOP 2021)

Abstract

Static analysis tools typically address the problem of excessive false positives by requiring programmers to explicitly annotate their code. However, when faced with incomplete annotations, many analysis tools are either too conservative, yielding false positives, or too optimistic, resulting in unsound analysis results. In order to flexibly and soundly deal with partially-annotated programs, we propose to build upon and adapt the gradual typing approach to abstract-interpretation-based program analyses. Specifically, we focus on null-pointer analysis and demonstrate that a gradual null-pointer analysis hits a sweet spot, by gracefully applying static analysis where possible and relying on dynamic checks where necessary for soundness. In addition to formalizing a gradual null-pointer analysis for a core imperative language, we build a prototype using the Infer static analysis framework, and present preliminary evidence that the gradual null-pointer analysis reduces false positives compared to two existing null-pointer checkers for Infer. Further, we discuss ways in which the gradualization approach used to derive the gradual analysis from its static counterpart can be extended to support more domains. This work thus provides a basis for future analysis tools that can smoothly navigate the tradeoff between human effort and run-time overhead to reduce the number of reported false positives.

Cite as

Sam Estep, Jenna Wise, Jonathan Aldrich, Éric Tanter, Johannes Bader, and Joshua Sunshine. Gradual Program Analysis for Null Pointers. In 35th European Conference on Object-Oriented Programming (ECOOP 2021). Leibniz International Proceedings in Informatics (LIPIcs), Volume 194, pp. 3:1-3:25, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2021)

Copy BibTex To Clipboard

@InProceedings{estep_et_al:LIPIcs.ECOOP.2021.3,
  author =	{Estep, Sam and Wise, Jenna and Aldrich, Jonathan and Tanter, \'{E}ric and Bader, Johannes and Sunshine, Joshua},
  title =	{{Gradual Program Analysis for Null Pointers}},
  booktitle =	{35th European Conference on Object-Oriented Programming (ECOOP 2021)},
  pages =	{3:1--3:25},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-190-0},
  ISSN =	{1868-8969},
  year =	{2021},
  volume =	{194},
  editor =	{M{\o}ller, Anders and Sridharan, Manu},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ECOOP.2021.3},
  URN =		{urn:nbn:de:0030-drops-140469},
  doi =		{10.4230/LIPIcs.ECOOP.2021.3},
  annote =	{Keywords: gradual typing, gradual verification, dataflow analysis}
}

4 Search Results for "Estep, Sam"

Rose: Composable Autodiff for the Interactive Web

Abstract

Cite as

Rose: Composable Autodiff for the Interactive Web (Artifact)

Abstract

Cite as

An Extensible User Interface for Lean 4

Abstract

Cite as

Gradual Program Analysis for Null Pointers

Abstract

Cite as

Thanks for your feedback!

Could not send message