Dr. Sundeep Prabhakar Chepuri (Assistant Professor at the Department of ECE at the Indian Institute of Science (IISc), Bangalore, India.)
Short Bio : Sundeep Prabhakar Chepuri received his M.Sc. degree (cum laude) in electrical engineering and Ph.D. degree (cum laude) from the Delft University of Technology, The Netherlands, in July 2011 and January 2016, respectively. He has held positions at Robert Bosch, India, during 2007-2009, Holst Centre/imec-nl, The Netherlands, during 2010-2011. He was a Postdoctoral researcher at the Delft University of Technology, The Netherlands, a visiting researcher at University of Minnesota, USA, and a visiting lecturer at Aalto University, Finland. Currently, he is an Assistant Professor at the Department of ECE at the Indian Institute of Science (IISc) in Bangalore, India.
Dr. Chepuri was a recipient of the Best Student Paper Award at the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) in 2015. He is currently the Associate Editor of the EURASIP Journal on Advances in Signal Processing (JASP) and a member of the EURASIP Signal Processing for Multisensor Systems’ Special Area Team. His research interests include mathematical signal processing, statistical inference and learning, applied to communication systems, network sciences, and computational imaging.
“Sparse Sampling and Learning on Graphs”
Abstract : Ubiquitous sensors generate prohibitively large datasets. Large volumes of data that we collect nowadays are complex in nature as they are collected on manifolds, irregular domains, networks, or point clouds. Extending classical signal processing concepts and tools to represent, interpret, and analyze signals defined on irregular graph domains is an emerging area of research known as graph signal processing.
In this talk, to begin with, we present near-optimal greedy methods to sparsely sample signals defined over irregular graph domains. We discuss how the underlying geometrical structure of the domain on which the data is defined can be exploited for sampling.
Most of the graph signal processing algorithms assume that the graph is given or can be appropriately defined. For scenarios where there is no initial graph available or we desire to modify a known graph as new data becomes available, we present methods to learn a sparse graph that best explains the available data. Using the observations recorded at a single node and a known excitation signal, algebraic algorithms to estimate the graph structure will be presented.