New research demonstrates that reformulating the mathematics underlying AI models could significantly reduce memory and storage demands on AI hardware.