New Generative Modeling Approach via 'Drifting' Achieves SOTA on ImageNet
A novel generative model using equilibrium-based objectives and a drifting mechanism posts state-of-the-art image generation results, offering a potential alternative to diffusion models.
A new preprint on generative modeling introduced an approach called "Generative Modeling via Drifting" that achieves state-of-the-art results on ImageNet, as flagged by @iScienceLuvr. The method uses an equilibrium-based objective rather than the denoising objective that underpins diffusion models.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.