Jonathan Ho

Jonathan Ho

An American researcher whose 2020 paper on Denoising Diffusion Probabilistic Models, written at UC Berkeley with Ajay Jain and Pieter Abbeel, simplified and stabilised Sohl-Dickstein's diffusion framework into a training objective that actually worked at scale, producing image quality that surpassed GANs. He followed it with classifier-free guidance, the mechanism that lets a text prompt steer the denoising process, which is what makes text-to-image systems controllable. (See also: Jascha Sohl-Dickstein, Generative AI, Synthetic media, Prompt engineering)