报 告 人：李攀
助理教授
Georgia Tech. ECE department
主 持 人：张牧涵 助理教授
必赢71886网址登录必赢626net入口
时 间：2024年8月14日 10:30-11:45
地 址：必赢71886网址登录资源西楼2201会议室
腾讯会议：185-749-537
报告题目：
Towards Efficient and Robust Graph and Geometric Machine Learning
报告摘要：
The application of Graph Machine Learning (GML) and Geometric Deep Learning (GDL) to enhance prediction capabilities for graph-structured data and point cloud data is prevalent in scientific disciplines. However, these applications often present fundamental challenges in model computation and generalization due to the ultra-large data scale and unstable data distribution. Specifically, applications in particle physics require processing point clouds on the scale of 10,000 points with a latency requirement of O(ms). Moreover, in scientific research, the data used for model training often comes from thoroughly investigated regimes, whose distributions frequently do not align well with the under-explored regimes, though only the latter attract research interest. Additionally, the mutual dependence of entities in a graph or point cloud breaks the assumptions adopted by most previous works, necessitating new problem formulations and principled methodologies. In this talk, I will introduce our recent studies addressing these problems. Some relevant papers are:
Locality-Sensitive Hashing-Based Efficient Point Transformer with Applications in High-Energy Physics, Miao et al., ICML 2024
Structural Re-weighting Improves Graph Domain Adaptation, Liu et al., ICML 2023
Pairwise Alignment Improves Graph Domain Adaptation, Liu et al., ICML 2024
报告人简介：
Pan Li has joined Georgia Tech. ECE department as an assistant professor since 2023 Fall, and holds an adjunct position at Georgia Tech. CSE and Purdue CS. Pan's research interest lies broadly in the area of machine learning and optimization with graph data. Pan Li's work has been recognized by several awards including NSF Early Career Award, several industry research awards, and the Best Paper award at the Learning on Graph Conference 2022.
