Scalable Sensing and Fast Learning using Sparsity and Codes
KEC 1001 Wed, 03/30/2016 - 9:00am Xiao (Simon) Li Postdoctoral Research Scholar, Electrical Engineering and Computer Science, University of California, Berkeley Abstract: Advances in pervasive sensing and ubiquitous communications have led to an era of data deluge, which imposes critical bottlenecks in terms of data size, processing speed and reliability. Sparsity in data structure offers promise in addressing this challenge of “scale", as evidenced by the success of fields like compressed sensing and sparse learning. However, existing popular methods are typically based on optimization techniques, which scale polynomially with the problem dimension, and are getting increasingly hard to scale. In this talk, I will view a wide range of high-dimensional sensing and learning problems through a novel coding-theoretic lens. Using a simple divide-and-conquer philosophy with modern tools from sparse estimation and coding theory, I will show how to enable scalable acquisition and fast computations that scale sub-linearly with the problem dimension. As motivating examples, I will describe how this coding-theoretic framework can be used for a host of problems from compressed sensing, wireless sensing to machine learning and data analytics. This helps us break the complexity barrier while providing strict performance guarantees, and has the potential to enable "real-time" processing of massive datasets in these applications. Bio: URL: http://eecs.oregonstate.edu/colloquium/scalable-sensing-and-fast-learning-using-sparsity-and-codes _______________________________________________ Colloquium mailing list [email protected] https://secure.engr.oregonstate.edu/mailman/listinfo/colloquium
