Content

(L-R) W. Richard Adrion, Mohammad Hajiesmaili, Neena Thota, Prashant Shenoy, Ramesh Sitaraman, David Irwin, Jimi Oke
(L-R) W. Richard Adrion, Mohammad Hajiesmaili, Neena Thota, Prashant Shenoy, Ramesh Sitaraman, David Irwin, Jimi Oke

Artificial intelligence (AI), algorithms and software—meet the next tool kit for creating our sustainable future. University of Massachusetts Amherst has been awarded $12 million over five years by the U.S. National Science Foundation to develop the field of computational decarbonization, or CoDec, a new branch of computer science and engineering that applies data-driven approaches to automate decarbonization across the electrical grid, the built environment, transportation and even computing itself.

“The crux of this project is to reduce the carbon footprint of societal infrastructure,” says Prashant Shenoy, Distinguished Professor and associate dean for the Manning College of Information and Computer Sciences (CICS) and principal investigator for this research initiative. 

For decades, researchers have focused on energy efficiency as a carbon-shrinking strategy, but Shenoy says we may have reached a tipping point where we need to emphasize carbon efficiency. As opposed to energy efficiency, carbon efficiency aims to reduce the amount of emissions—not just energy—per unit of work done.

The obvious solution is to switch from fossil fuels to renewables like solar and wind. The challenge is that solar panels only produce when it’s sunny out, windmills only turn when there’s wind and these factors are highly dependent on where you are and what time of day it is.

CoDec aims to use computer science tools to automate, coordinate and maximize carbon efficiency based on this time and location puzzle across four domains of infrastructure: computing, electricity, buildings and transportation. The project will then create software interfaces to optimize our most carbon-intensive activities with the greenest energy possible. 

“The different infrastructures we’re looking at in this project all have different dimensions of flexibility in terms of time and space,” says David Irwin, professor of electrical and computer engineering and one of the UMass Amherst contributors on this project. “For example, the heating in a building can’t be shifted in space—the building can’t move. However, computing is uniquely flexible across time, space and performance. You can move running a computing job or serving a web page from New York to California. And you can do it very quickly. But all of these infrastructure systems have some flexibility that we’ll be looking to exploit.”

The researchers have organized their work into a sense-optimize-reduce framework, commonly used in cyber-physical systems. Sense provides transparency, defining carbon footprints across the different infrastructure domains. Next, the researchers will design algorithms, AI and other computational methods to optimize a system’s carbon efficiency based on where and when the green energy is available. Finally, the reduce phase builds the software that interfaces with the four domains so that these optimization strategies can be implemented in the real world.

Here, Shenoy highlights the importance of AI. On one hand, AI is a powerful tool that can find optimization solutions. “However, it is extremely energy- and carbon-intensive to train large AI models like ChatGPT,” Shenoy says. “So we are going to think carefully about how to make the AI workloads running in data centers more carbon efficient.”

One of the key elements of this initiative is understanding the interconnected nature between these seemingly disparate infrastructure domains. Consider working from home: Yes, it shrinks a person’s carbon footprint from transportation, but it may simply shift the carbon burden to their home energy use. 

Shenoy explains that this is called the interdependency gap. “There are these subtle couplings between sectors,” he says. “There are decarbonization efforts already in progress, but they are siloed. People have looked at, ‘how do I do decarbonize computing?’ Or ‘how do I do this for transportation?’ But not, ‘How do I do this for society?’”

To further complicate things, this leads to tension between these dependencies. “It’s not always clear which one should take priority when you cannot optimize them all,” he adds.

Irwin elaborates on this human part of the equation: “If the performance from the website goes down a little bit, that might be okay, but if I can’t keep my home warm, that’s not okay. The system is going to have to be more flexible to handle the increase in demand and prioritize different uses of energy.”

Another unknown the researchers highlight is the scale gap. “People in computing are often worried about milliseconds of response time,” says Irwin. “But people haven’t looked at larger time scales.” However, optimizing for carbon efficiency needs to be considered over months, years or longer (which the researchers refer to as “the mesoscale”) so establishing these long-term computing techniques is another goal of the CoDec project.

To address these multifaceted research questions, the CoDec project will develop theoretical foundations of computational decarbonization by combining safety-critical optimization and data-driven learning techniques. The researchers will also develop new courses in this emerging field and work with a group of industry partners to transition decarbonization research to practice.

The CoDec initiative is one of three Expeditions projects receiving a portion of $36 million from the NSF aimed at addressing grand challenge problems in computing.

“We are proud that NSF has chosen UMass Amherst to lead the development of revolutionary new technologies to assist in the fight against climate change,” says Laura Haas, Donna M. and Robert J. Manning Dean of CICS. “Led by Prashant Shenoy, whose previous work in distributed systems and sustainable computing  has made him a preeminent researcher in this area, the team brings a wealth of expertise to the biggest challenge of our time, literally computing for the common good.”

This holistic endeavor can only be achieved with an interdisciplinary team across multiple institutions. 

“The most complex and vexing problems facing our planet, such as decarbonizing the high-performance computing and data centers that drive today’s digital economy, require interdisciplinary teams that cut across computing systems, engineering, economics and the built environment,” says Sanjay Raman, dean of the College of Engineering and professor of electrical and computer engineering. “UMass Amherst is committed to building and sustaining such teams to solve these global grand challenges, and training the next generation of scientists and engineers with an interdisciplinary mindset in their DNA.”

From UMass Amherst, other collaborators include assistant professor Mohammad Hajiesmaili, Ramesh Sitaraman, Distinguished Professor and associate dean of educational programs and teaching; Neena Thota, associate chair of teaching development and senior teaching faculty; and W. Richard Adrion, professor emeritus from CICS; as well as Jimi Oke, assistant professor of civil and environmental engineering.

The other research institutions involved in the U.S. NSF Expeditions in Computing for Computational Decarbonization of Societal Infrastructures at Mesoscales are Carnegie Mellon University, Massachusetts Institute of Technology, the University of Chicago, the University of California Los Angeles and the University of Wisconsin.

Article posted in Research for Faculty , Staff , Prospective students , Current students , Alumni , and Public