Seminar: New tensor algebra for learning on dynamic graphs
Speakers: Shashanka Ubaru and Lior Horesh, IBM T. J. Watson Research Center
Host: Abram Magner (firstname.lastname@example.org)
Time: Thursday, May 6, 2:00 p.m. Eastern time.
Meeting ID: 957 2371 6801
One tap mobile +16465588656,,95723716801# US (New York) +13126266799,,95723716801# US (Chicago)
Title: Into another dimension - a new tensor algebra for learning on dynamic graphs
Abstract: In recent years, a variety of graph neural networks (GNNs) have been successfully applied for representation learning and prediction on graphs. In many of the real-world applications, the underlying graph changes over time, however, most of the existing GNNs are inadequate for handling such dynamic graphs. In this study, we explore a novel technique for learning embeddings of dynamic graphs using a tensor algebra framework. The tensor framework is based upon the notion of tensor-tensor product, an algebraic formalism to multiply tensors, which inherits mimetic matrix properties. We show that the proposed tensor formalism is a natural extension of popular graph convolutional network (GCN) to the dynamic graph setting. We present numerical experiments on real-world datasets to demonstrate the intrinsic advantages of the proposed tensorial architecture for prediction tasks on dynamic graphs. We also consider an application related to the COVID-19 pandemic, and show how the tensor framework can be used for early infection detection and uncertainty quantification in testing from contact tracing data.