Rose: Composable Autodiff for the Interactive Web
This program is tentative and subject to change.
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 are unperformant 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 compose 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.
This program is tentative and subject to change.
Tue 17 SepDisplayed time zone: Amsterdam, Berlin, Bern, Rome, Stockholm, Vienna change
10:30 - 12:00 | |||
10:30 15mTalk | Defining Name Accessibility using Scope Graphs Technical Papers Link to publication Pre-print | ||
10:45 15mTalk | Rose: Composable Autodiff for the Interactive Web Technical Papers Sam Estep Carnegie Mellon University, Wode Ni Carnegie Mellon University, Raven Rothkopf Barnard College, Joshua Sunshine Carnegie Mellon University | ||
11:00 15mTalk | Failure Transparency in Stateful Dataflow Systems Technical Papers Aleksey Veresov KTH Royal Institute of Technology, Jonas Spenger KTH Royal Institute of Technology, Paris Carbone KTH Royal Institute of Technology, Philipp Haller KTH Royal Institute of Technology Pre-print | ||
11:15 15mTalk | Fair join pattern matching for actors Technical Papers Philipp Haller KTH Royal Institute of Technology, Ayman Hussein Technical University of Denmark, Hernan Melgratti University of Buenos Aires, Argentina, Alceste Scalas Technical University of Denmark, Emilio Tuosto Gran Sasso Science Institute, L'Aquila, Italy | ||
11:30 15mTalk | Constrictor: Immutability as a Design Concept Technical Papers Pre-print | ||
11:45 15mTalk | A Language-Based Version Control System for Python Technical Papers Luís Carvalho NOVA School of Science and Technology, João Costa Seco NOVA-LINCS; Nova University of Lisbon |