`

Timezone: »

 
Gotta Go Fast with Score-Based Generative Models
Alexia Jolicoeur-Martineau · Ke Li · Rémi Piché-Taillefer · Tal Kachman · Ioannis Mitliagkas

Tue Dec 14 06:45 AM -- 07:30 AM (PST) @ None
in The Symbiosis of Deep Learning and Differential Equations »
Event URL: https://openreview.net/forum?id=gEoVDSASC2h »

Score-based (denoising diffusion) generative models have recently gained a lot of success in generating realistic and diverse data. These approaches define a forward diffusion process for transforming data to noise and generate data by reversing it. Unfortunately, current score-based models generate data very slowly due to the sheer number of score network evaluations required by numerical SDE solvers. In this work, we aim to accelerate this process by devising a more efficient SDE solver. Our solver requires only two score function evaluations per step, rarely rejects samples, and leads to high-quality samples. Our approach generates data 2 to 10 times faster than EM while achieving better or equal sample quality. For high-resolution images, our method leads to significantly higher quality samples than all other methods tested. Our SDE solver has the benefit of requiring no step size tuning.

Author Information

Alexia Jolicoeur-Martineau (Lady Davis Institute)

I am looking for opportunities in AI research (not applied data analysis, I'm only interested in research). My main interests are generative models and deep learning in general. I have strong research experience, I have more than 10 publications under belt, 4 of them first-author.

Ke Li (Google / SFU)
Rémi Piché-Taillefer (Mila)
Tal Kachman (Radboud University)
Ioannis Mitliagkas (University of Montreal)

More from the Same Authors