Geometric Multi-scale Low Dimensional Representation of High Dimensional Signals
Dr. Sang (Peter) Chin, Professor of Electrical and Computer Engineering, Johns Hopkins University
Thursday, May 2, 2013, 4:15pm
This talk will focus on our attempts for developing techniques for efficiently representing high dimensional data, extracting their features and understanding their intrinsic low-dimensional information. Low-rank matrix approximation is one such attempt, as it proved to be an efficient way of representing the signal sparsity in the principal component domain. This is an intimate connection to compressive sensing and provides a robust alternative framework to recover low-dimensional structures from high-dimensional observations, especially for scenarios where the data is highly incomplete or severely damaged. These low-rank models have been beneficial in solving a number of applications in various diverse disciplines, from signal processing to pattern recognition, machine learning and computer vision. Furthermore, we have been als developing more general non-linear multi-scale techniques for analyzing the geometry of high-dimensional point clouds, which have low-intrinsic dimensionality but are corrupted by a high-dimensional noise. Such point clouds, or associated graphs encoding relationships between points, are at the core of many machine learning, feature extraction, and data analysis algorithms, and therefore the development of effective analytical techniques for these data is of crucial importance. I will describe some recent results and future directions in this area and describe a few applications in tracking vehicles through persistent E/O images, detection and classification in hyper-spectral images (HSI), face recognition, and finding a mysterious node in social networks.
About the Speaker
Dr. Chin is currently a chief scientist of Cyber Technologies Branch at Applied Physics Laboratory and a research professor in the Dept. of Electrical and Computer Engineering of Johns Hopkins University, where he is leading multiple research thrusts in the areas of game theory, geometric machine learning theory, compressive sensing, extremal graph theory, cognitive radio and cyber security supported by grants from NSF, ONR, AFOSR and OSD. He currently serves as chair of the annual SPIE conference on cyber sensing, a member of Computational cyber security in Compromised Environments (C3E) Consortium, and visiting fellow at London Institute of Mathematical Sciences. He received his PhD in mathematics from MIT and is a Phi Beta Kappa graduate of Duke University where he was a triple major in electrical engineering, computer science and mathematics.
For more information, contact Louise Cullen at 603-646-8794 or Louise.A.Cullen@Dartmouth.edu.