Researchers develop unprecedented tool to ease a major AI pain point: ‘Particularly valuable’

The rise in deep learning/AI models goes hand in hand with increased energy consumption and planet-warming pollution, sparking North Carolina State University researchers to develop a method of predicting those costs to make informed decisions.

Jung-Eun Kim, an assistant professor of computer science at the educational institution, authored a paper on the subject explaining the reasoning behind the work, according to an article shared by TechXplore.

“If we want to address sustainability issues related to deep learning AI, we must look at computational and energy costs across a model’s entire life cycle – including the costs associated with updates,” Kim explained.

“If you cannot predict what the costs will be ahead of time, it is impossible to engage in the type of planning that makes sustainability efforts possible. That makes our work here particularly valuable.”

People are investing heavily in the advancement of artificial intelligence, and the tech could be applied to solving environmental issues. It’s already being used to map the destructive nature of sand dredging and track methane pollution that builds up in the atmosphere and heats up the planet, according to the United Nations.

However, data centers that house those systems require an enormous amount of energy, from powering the process to keeping those machines cool enough to function.

Watch now: Giant snails invading New York City?

Using clean energy to power the centers is part of the solution while developing more efficient cooling systems and reducing the amount of electronic waste generated by these installations are also key in making these processes more sustainable.

The Department of Energy has found that data center load growth has tripled in the last ten years and could potentially triple again by 2028. In fact, data centers in the U.S. consumed around 4.4% of the total electricity used in the country in 2023, with that number estimated to grow to up to 12% in the next five years.

Training deep learning models is a computationally intensive process, as the NC State report explained, and scheduling necessary updates for the most efficient time is desirable.

“Regardless of what is driving the need for an update, it is extremely useful for AI practitioners to have a realistic estimate of the computational demand that will be required for the update,” said Kim.

“This can help them make informed decisions about when to conduct the update, as well as how much computational demand they will need to budget for the update.”

The technique they developed is called the REpresentation Shift QUantifying Estimator (RESQUE), which compares both the current and updated datasets in order to estimate the computational and energy costs involved.

“In the bigger picture, this work offers a deeper understanding of the costs associated with deep learning models across their entire life cycle, which can help us make informed decisions related to the sustainability of the models and how they are used,” Kim summarized.

“Because if we want AI to be viable and useful, these models must be not only dynamic but sustainable.”

Join our free newsletter for weekly updates on the latest innovations improving our lives and shaping our future, and don’t miss this cool list of easy ways to help yourself while helping the planet.