Contemporary generative models mainly utilize diffusion models to create realistic high-resolution samples. Nevertheless, diffusion models necessitate an iterative denoising procedure for generating samples, which currently hinders their widespread application in real-time industrial settings. In this presentation, I will discuss the diffusion model family that enables the few-step sample generation, including distillation methods.

I got Ph.D. at KAIST in 2023, advised by Professor Il-Chul Moon, with the topic of generative models. I am currently working with SONY and I will go Stanford University in Professor Stefano Ermon's group as a postdoctoral researcher next spring.