Thursday, April 11th, 2024
11:00 AM - 11:50 AM
Salisbury Labs 011
Title: Diffusion Generative Models
Abstract: Diffusion generative models have found increasing success in generating synthetic data, including highly realistic images and videos. In this talk we will see how one can use tools from stochastic differential equations to construct and train diffusion generative models, and how diffusion models can be applied to a wide variety of tasks such as image generation, image resynthesis, denoising, and molecular drug discovery. We also discuss how diffusion models can be combined with Large Language Models such as ChatGPT to provide state-of-the art image generation capabilities, conditional on a given text prompt. A basic knowledge of continuous-time stochastic processes is assumed.
