I am honored to have the opportunity to talk about our work on diffusion, control, inference, and learning. i) Stochastic diffusion is a handy tool. ii) control for learning is promising. There is plenty control theory can offer for machine learning tasks Machine Learning at Georgia Tech GTaerospace

Building generative models for large-scale content with only small pieces of data! 🚀 👇chat with Jiaming Song #187 #CVPR2023 #diffusion

Exhibit Hall 1 #211👇 would like to discuss generative models and scaling laws

An incredible journey with an amazing team to develop a powerful and ‘safer’ image generator #MachineLearning #AI

(2/n) “Improved dimension dependence of a proximal algorithm for sampling. COLT 2023” We apply this Gaussian concentration inequality to the analysis of the proximal sampler and get state-of-the-art sampling convergence results wrt dimension dependence under various conditions.

concatenated attention is secretly a guidance in diffusion model