Search
Generative artificial intelligence (AIGC) is making rapid strides in diverse fields, including text-to-image and text-to-video generation, embodied intelligence, and world modeling. Generative algorithms, with diffusion models as a prime example, are emerging as pivotal technologies for mimicking human imagination. Yet, existing diffusion models encounter challenges such as sluggish performance and excessive energy consumption when deployed on conventional digital computing platforms. This stems from the von Neumann architecture's separation of memory and computing units, which necessitates extensive data shuttling. Furthermore, diffusion models hinge on solving neural differential equations, compelling digital computers to discretize the continuous generation process, thereby escalating both time and energy requirements.
-
C114 Communication Network -
Communication Home
