HomeNewsListProfessor Yuejie Chi visited on June 26th

Professor Yuejie Chi visited on June 26th

Professor Yuejie Chi from the Electrical and Computer Engineering Department, the Ohio State University visited our lab on June 26th, 2015. She gave a talk on "Tracking and Sketching of Covariance Structures For High-dimensional Streaming Data" in Room 8-208, Rohm Building at 10:30.  

Fig 1. Professor Chi is giving the talk

Fig 2. Professor Chi is answering questions

Dr. Yuejie Chi is an assistant professor in the Electrical and Computer Engineering Department at The Ohio State University, with a joint appointment in the Biomedical Informatics Department at the Wexner Medical School since September 2012. She received a M.A. and a Ph.D. in Electrical Engineering from Princeton University in 2009 and 2012 respectively, and a B.Eng. in Electrical Engineering from Tsinghua University, China in 2007. She is the recipient of the Young Investigator Program Award from Air Force of Scientific Research and Office of Naval Research, respectively in 2015, the Ralph E. Powe Junior Faculty Research Award in 2014, IEEE Signal Processing Society Young Author Best Paper Award in 2013 and the Best Paper Award at the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) in 2012. Her research interests include high-dimensional data analysis, statistical signal processing, machine learning and their applications in network inference, active sensing, image analysis and bioinformatics.



Tracking and Sketching of Covariance Structures For High-dimensional Streaming Data


The explosion of high-dimensional and high-rate data streams has overwhelmed the computational and storage power of traditional sensor suites. On one hand, the data samples may suffer from missing data, so that only a small subset of its entries are observed, for example in online recommendation systems; on the other hand, the sensors may be power-hungry and resource-limited that can only allow one pass of the data samples at each instance. In this talk, we demonstrate how exploiting low-dimensional geometry of the data stream allows efficient extraction and tracking of its covariance structure with a reduction of complexity in acquisition, storage and computational cost. We start by describing an algorithm called PETRELS that can track the data stream from its partial observation in a timely fashion by assuming it lies in a low-dimensional subspace. We then describe two sketching schemes to recover the covariance matrix of a data stream with a single quadratic (energy) sketch of each data sample, without observing the original data stream. The first scheme is based on energy aggregations, and the second scheme further reduces the energy aggregations into pairwise bit comparisons. Rigorous theoretical guarantees are established to demonstrate the near-optimal performance of the proposed sketching schemes. Numerical examples on direction-of-arrival estimation are presented to corroborate the theoretical findings.