Researchers at the University of Michigan have developed an open-source optimization framework called Zeus that could reduce the energy demands of training deep learning models by up to 75 percent without the need for new hardware and with only minor impacts on the time it takes to train a model. The framework was presented at the 2023 USENIX Symposium on Networked Systems Design and Implementation (NSDI) in Boston. Deep learning is a subset of machine learning that relies on multilayered models to power a range of applications from image-generation models and expressive chatbots to recommender systems. The increased climate burden from artificial intelligence is a pressing concern and the new framework could help reduce energy consumption figures.
Related Posts
