Abstract:
How do we quickly detect small solar flares from a large volume of videos captured by NASA satellites? How should we represent high-dimensional data when it is slowly time-varying? High-dimensional data often exhibits low-dimensional structures, including signal sparsity under a given basis, or the data may lie near a manifold.

In this talk, Dr. Yao Xie will present answers to these questions by exploiting various low-dimensional structures and by finding ways to connect modern signal processing (such as compressed sensing) with traditional statistics (such as change-point detection). In particular, Dr. Xie will focus on multi-sensor change-point detection, which utilizes signal sparsity to achieve much quicker detection, and the MOUSSE algorithm that forms online estimates of the dynamic manifold, which enables high-dimensional change-point detection with missing data. She establishes theoretical performance guarantees for these methods by using old and new tools from sequential analysis, compressed sensing, geometric wavelet, subspace tracking, and information theory. She will also mention other applications such as low-complexity multi-user detection in large wireless networks and early breast cancer detection.

Learn more about solar flares in this 2-minute video.  

Speaker Bio:
Yao Xie joined Georgia Tech's H. Milton Stewart School of Industrial and Systems Engineering in August 2011 as an assistant professor. Prior to this, she was a research scientist in the Electrical and Computer Engineering Department at Duke University. She received her Ph.D. degree in electrical engineering (minor in mathematics) from Stanford University in 2011. Her research interests include sequential statistical methods, statistical signal processing, big data analysis, compressed sensing, and machine learning with applications in wireless communications, sensor and social networks, and medical and astronomical imaging.