Deep Learning Progress Report

Abstract

This progress report presentation involves two talks. In the first talk, we report the ongoing progress in exploring modern generative approaches to conditional distribution estimation, with a particular focus on the denoising diffusion probabilistic model (DDPM). By learning to reverse a stochastic noising process, DDPM provides a powerful framework for sampling from complex conditional distribution and serves as a promising tool for probabilistic modeling. In the second talk, we present progress on another three ongoing projects. The first introduces a deep learning method for functional data prediction. The second develops a generative modeling framework for synthesizing experimental design matrices. The third offers an introductory tutorial on flow matching, a flexible deep learning framework for building generative models. We conclude by outlining a future direction that explores the use of AI agents in coding and scientific research.

線上視訊請點選連結