Nonparametric anomaly detection on time series of graphs
University of Texas at Dallas
Date: Wednesday, October 21, 2020
Time: 4:00 pm - 5:00 pm
Abstract: Anomaly detection is one of the primary targets in the analysis of dynamic networks. Its applications range from tracking new gang formation and money laundering schemes to identifying neurological disease progression to resilience analysis of power grids. However, most currently available methods for anomaly detection focus on lower order graph structures (i.e., connectivity features at the level of individual nodes and edges) and disregard the temporal dependence among adjacent network time snapshots. In this talk, we address these challenges by introducing a new distribution-free anomaly detection method for time evolving networks, based on generalized spectrum of a motif tensor. We evaluate the new procedure on synthetic data and real-world case studies. Our numerical studies indicate competitive performance of the new method in detecting higher order structural changes in time evolving networks.
*All talks will be held virtually through Zoom. Please contact firstname.lastname@example.org for talk links.*