Alan Edelman
Massachusetts Institute of Technology, MA·Computer Science
Computer simulations predict weather, floods, air pollution and much more. Building and running them, however, often takes deep programming expertise and expensive supercomputers.
- Fields
Computer Science
Mathematics & Statistics
Earth & Environmental Science
- Topics
scientific computing
automatic differentiation
Enzyme
- Grants
- 3 known grants, through Oct 2027
What they work on
- Build an equation-based geoscientific modeling system in which a compiler automatically turns systems of equations into computer models, adding components prioritized at community workshops and documenting them on a dedicated website.
- Set up decentralized governance and interactive educational experiences so the equation-based geoscience modeling network can grow into a user-supported community over the long term.
- Develop machine-learned surrogate models, fast stand-ins for slow code, for atmospheric chemistry in air pollution models, along with improved dimensionality reduction methods that shrink these chemical systems to fewer variables.
- Apply the same machine-learning methods to model wildfire plume rise, how high smoke is lofted, which is poorly captured in current models but key to predicting smoke transport.
- Create a modular, differentiable, equation-based platform for atmospheric chemical transport modeling that lets physics-based and machine-learning model components be combined.
- Develop Enzyme, a tool that automatically generates derivatives of programs in LLVM-based languages such as C/C++, Fortran, Julia and Rust, into production-quality software integrated with vendor compilers and maintained by an open-source community.
More about the research
The projects here aim to remove those barriers. One goal is to let scientists write down the equations that describe a system and have software produce fast, accurate code automatically. A second is to make that code "differentiable," meaning the software can compute how outputs change as inputs change. Differentiability lets machine learning be combined with physics-based models. Across the three collaborative grants, the work spans equation-based modeling for the geosciences, machine-learning shortcuts for air pollution models, and general tools for automatic differentiation.
Why it matters
Faster, easier-to-use simulation tools could let more scientists, educators and policymakers run models of weather, water supply, air pollution, floods and tsunamis without supercomputers or specialist programming skills. Better models of wildfire plume rise could improve forecasts of where smoke travels, an increasingly important public health issue. Automatic differentiation for existing code could let engineers in many fields apply neural networks without rewriting their applications.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Collaborative Research: CAIG: Interpretable, Stable, Mass-Conserving AI for Air Pollution Modeling
NSF · Nov 2024 – Oct 2027
Collaborative Research: GEO OSE Track 1: Building an Equation-Based Geoscientific Modeling Network
NSF · Sep 2025 – Aug 2027
Collaborative Research: Research Infrastructure: CIRC: ENS: Bridging Science and AI Through Differentiable Program Transformations
NSF · Jul 2024 – Jun 2027
Closest labs
Sri Hari Krishn Narayanan
University of Chicago, IL·Computer Science
Develop an open-source, general-purpose automatic differentiation framework in Julia that computes exact derivatives of program code for both large-scale Earth system models and scientific machine learning methods.
Patrick Heimbach
University of Texas at Austin, TX·Earth & Environmental Science
Set deep-ocean observing requirements by working with Global Ocean Observing System expert panels to refine Essential Ocean Variables, and apply them through regional demonstration projects.
Cheng-Hsuan Lu
University at Albany, NY·Earth & Environmental Science
Integrate the MUSICA chemistry and aerosol library into the MPAS-A global weather model to create CheMPAS-A, which simulates atmospheric chemistry seamlessly from global to local scales.
The school
Massachusetts Institute of Technology: 25 labs on PI Pages, $1.1B a year on research, 692 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.