Communications of the ACM
Reducing the Carbon Footprint of AI
A new automated artificial intelligence system for training and running certain types of neural networks has a relatively small carbon footprint.
Credit: MIT News
Massachusetts Institute of Technology (MIT) researchers have developed an automated artificial intelligence (AI) system for training and running certain types of neural networks, which has a relatively small carbon footprint.
The researchers built the system via automated machine learning, which trained the OFA network.
Using the system to train a computer-vision model, the researchers calculated that the effort required approximately 1/1,300th the carbon emissions of modern state-of-the-art neural architecture search strategies, and reduced the duration of inference by 1.5 to 2.6 times.
MIT's John Cohn said, "The upside of developing methods to make AI models smaller and more efficient is that the models may also perform better."
From MIT News
View Full Article
Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA
No entries found