Speaker
Yusu Wang (UC San Diego)
Title
Infusing topological information to machine learning
Abstract
Topological data analysis has received significant development in the past decades, building upon the advancement in applied and computational topology. In recent years, a range of approaches have been proposed to combine topological data analysis with machine learning pipelines. In particular, it can sometimes be challenging for machine learning frameworks to capture or respect global structure of data, an aspect that topological methods can be good at. In this talk, I will describe some of our recent work in infusing topological information into machine learning (especially neural networks) to further augment machine learning approaches. I will in particular introduce our work on persistent-enhanced graph neural networks, as well as using topological structures beyond persistence to augment neural networks. This talk is based on several pieces of collaborative work with various collaborators, whose names I will give during the talk.
Contact
tes-summer2021@math.tu-berlin.de