Strubell Among CMU Researchers To Tackle Carbon Use, Sustainability Through NSF Expeditions in Computing Awards

The LTI's Emma Strubell is among a team of researchers contributing to two NSF multi-institution research initiatives aimed at reducing the use of carbon and creating sustainable computing.

Researchers from Carnegie Mellon University's School of Computer Science will contribute to two multi-institution research initiatives aimed at reducing the use of carbon and creating sustainable computing.

The projects recently received funding through the U.S. National Science Foundation's (NSF) Expeditions in Computing Awards program, which is providing $36 million to three projects selected for their potential to revolutionize computing and make significant impacts toward reducing the carbon footprint of the lifecycle of computers. The Expeditions program is an ambitious initiative providing some of the largest investments from the NSF's Directorate for Computer and Information Science and Engineering (CISE) to support transformative research poised to yield lasting impacts on society, the economy and technological advancement.

"We are thrilled to announce these visionary projects that will advance environmental responsibility and foster innovation in the field of computing," said Dilma Da Silva, acting assistant director for CISE. "Congratulations to these pioneering teams, whose research will forge new pathways in computational decarbonization and in revolutionizing operating system design with machine learning."

Computational Decarbonization

One project, NSF Expeditions in Computing for Computational Decarbonization of Societal Infrastructures at Mesoscales, will develop the new field of computational decarbonization (CoDec) to focus on the lifecycle of carbon emissions — from production and use — across different societal infrastructures such as computing, transportation, buildings and the power grid.

Climate change from carbon and other greenhouse gas emissions poses an existential threat to the Earth's ecosystem and humanity. But curbing energy use and focusing on energy efficiency will not sufficiently reduce carbon emissions. Instead, efforts should aim for decarbonization by focusing on and optimizing for carbon efficiency.

SCS faculty will collaborate with researchers from five institutions to develop new sensing technologies to better show and track carbon footprints, use AI and optimization theories to further energy flexibility, and ultimately create software and systems to optimize carbon use. Yuvraj Agarwal will serve as the lead principal investigator from CMU and will be joined by Zico Kolter on the project team. Agarwal and Kolter bring a host of expertise to the project, in topics including sensing, systems, security and privacy, artificial intelligence, using data to incentivize decision making, and understanding how computing interacts with smart buildings and efficient infrastructure.

"This project will use computing techniques to help address decarbonization, which is arguably one of this generation's most pressing societal challenges," said Agarwal, an associate professor of computer science in the Software and Societal Systems Department (S3D) and the Human-Computer Interaction Institute. "A cross-disciplinary approach is needed to make progress, and this is exactly the type of challenge that CMU, its School of Computer Science and especially the Software and Societal Systems Department within the school is designed to tackle."

Agarwal will work to devise systems to track carbon in a verifiable manner at different scales and across societal infrastructures that we all use. This effort includes using sensing and inference techniques to characterize the embodied and operational carbon and presenting the information in an actionable format to spur change. Agarwal will create systems that treat carbon emissions as a first-class design constraint and optimization objective.

"Real-time data on carbon emissions from power plants and the energy grid; embedded sensors; and the connectivity of buildings, vehicles and infrastructure are giving us a better picture of the lifecycle of carbon and the data needed to develop methods to reduce it," Agarwal said.

Kolter will develop machine learning methods to integrate optimization processes and simulations directly within the models themselves. This process can allow the models to leverage the benefits of data-driven AI techniques, while simultaneously incorporating known knowledge about physical processes and other "constraints" of real-world environments.

"Continued advancements in computing and AI coupled with the severe impacts of climate change make the timing ripe for this project," said Kolter, an associate professor in the Computer Science Department.

Sustainable Computing

Emma Strubell, the Raj Reddy Assistant Professor of Computer Science in the Language Technologies Institute, will contribute to a second Expeditions project. NSF Expeditions in Computing: Carbon Connect – An Ecosystem for Sustainable Computing, co-led by Harvard University and the University of Pennsylvania, will lay the foundation for sustainable computing.

Limiting global warming in accordance with the Paris Agreement will require an estimated 45% reduction in greenhouse gas emissions from the information and communication technology sector by 2030 over a 2020 baseline. To achieve this goal, Carbon Connect will focus on reducing the environmental impact of computer systems themselves by identifying and developing the foundational technologies required to potentially spark a transformation in how these systems are manufactured, allocated and consumed. This work could lead to a more responsible and sustainable approach to advancing computing technologies.

"Recent technological advances, such as generative AI, have great potential to help address substantial societal challenges, including climate change. However, these same technologies come with a high carbon footprint that must be drastically reduced over a short time period to meet climate goals," said Strubell, who will lead the project's applications theme, which will focus on generative AI and virtual spaces. "Realizing the necessary reduction in emissions will require cross-cutting innovations that take into account inherent tradeoffs in emissions up and down the technology stack, alongside close collaborations with industry and policymakers, which is exactly what we aim to achieve with this highly collaborative project."

Beyond advancing the state of the art in hardware, software and datacenter efficiency, Carbon Connect intends to establish new standards for carbon accounting in the computing industry, thereby influencing future energy policy and legislation.

Learn more about the Expeditions in Computing Awards on the NSF's website.