 |
Center for Nonlinear Studies |
AbstractGraph algorithms are a standard tool for any data analysis task in which connections between data points matter. In this talk, I will cover a number of graph algorithms I have worked on in recent years in order to understand connections between nodes beyond pairwise connections. I will show you how I tackle these problems with a linear algebra hammer and explore the useful interplay between graph analysis and matrix computations. The graph problems I will cover are node ranking problems, link prediction, network alignment, and node embeddings. While talking about these problems I will explore two key algorithmic techniques that I often use: approximation methods and higher order methods. BioHuda Nassar is a postdoctoral fellow at Stanford University and obtained her PhD in Computer Science from Purdue University. While at Purdue, Huda's main focus was designing a new class of low rank algorithms to solve multiple variations of the network alignment problem. More generally, Huda's research interests in network science span a wide range of topics including prediction problems, higher order methods, and approximation algorithms. During her PhD, Huda received the John R. Rice Fellowship for contributions to scientific computing and in 2020, she was nominated to the Rising Stars in Computational and Data Science. In addition to her scientific research, Huda is an enthusiastic Julia user and the author of the graph algorithms package MatrixNetworks.jl.
Host: Anatoly Zlotnik