Physicists at the Massachusetts Institute of Technology (MIT) are exploring how principles from physics can enhance artificial intelligence (AI) technology. Their research focuses on generative AI models, which use equations of physical processes rather than black box algorithms. One promising approach is the Poisson flow generative model (PFGM), which represents data using charged particles and an electric field. The PFGM can create high-quality images as quickly as diffusion-based models, and an upgraded version called PFGM++ offers greater variability. The researchers are also investigating other physical processes to develop new families of generative models with potential applications beyond image generation.
https://www.quantamagazine.org/new-physics-inspired-generative-ai-exceeds-expectations-20230919/