Summary: Prof. Romit Maulik presents anomaly detection algorithms for detecting intermittent events in dynamical systems proposed in his recently published research article - “Online data-driven changepoint detection for high-dimensional dynamical systems”. In this work, data-driven anomaly detection algorithms are devised for high-dimensional dynamical systems that exhibit intermittent events. The proposed algorithm addresses (a) high-dimensionality through deep learning compression via autoencoders and, (b) online detection of changepoints via a conjugate Bayesian formulation. The proposed algorithms are tested on prototypical dynamical systems given by the (a) Lorenz-63 system, (b) the Rossler system, and (c) a high-dimensional forced Kolmogorov flow. Further analysis shows that the proposed method is able to detect transitions that are associated with visiting new regions of phase space during the evolution of the dynamics.
About the Speaker: Prof. Romit Maulik is an assistant professor in the Information Science and Technology Department at Pennsylvania State University and a faculty affiliate at the Mathematics and Computer Science Division at Argonne National Laboratory. He was the Margaret Butler Postdoctoral Fellow at Argonne National Laboratory. He holds a doctoral degree in mechanical and aerospace engineering from Oklahoma State University. Dr. Maulik leads the research team at the Interdisciplinary Scientific Computing Laboratory (ISCL), where he focuses on leveraging the concepts of applied mathematics, physics, and computer and data science to design computational strategies for multidisciplinary engineering applications.
Event Link
