CSL Communications Group Calendar
http://illinois.edu/calendar/list/3123
CSL Communications Group CalendarCSL SEMINAR - “Second Order Estimation for High-Dimensional Time Series: Covariance and Precision Matrices”
http://illinois.edu/calendar/detail/3123/32092430
http://illinois.edu/calendar/detail/3123/32092430Mon, 22 Sep 2014 16:00:00 CDT**Abstract**
Covariance matrix and its inverse (a.k.a. precision matrix) play the fundamental roles in statistics, signal processing, communications, and machine learning. With the advance of modern data acquisition technology, huge volumes of data can be collected in the time domain. As an important associated research problem, one should extract information from a large number of variables in the presence of subtle interactions between the spatial and temporal dimensions. In this talk, we consider the structured covariance matrix estimation and related problems for high-dimensional times series. Using the functional dependence measure, we obtain the rate of convergence for the thresholded estimate and illustrate how the dependence affects the rate of convergence. Asymptotic properties are also obtained for estimators of the precision matrix and their functionals. Our theory substantially generalizes earlier works by allowing dependence, by allowing nonstationarity, and by relaxing associated moment conditions.
******Bio******
Xiaohui Chen is an Assistant Professor of Statistics at UIUC since August 2013. He received the PhD degree in Electrical and Computer Engineering from the University of British Columbia (UBC) in 2012. He earned the BSc from Zhejiang University in 2006 and the MSc from UBC in 2008. He was a post-doctoral researcher at the Toyota Technological Institute at Chicago (TTIC). His current research interests includes the broad fields of high-dimensional statistics, time series analysis, statistical machine learning and signal processing.CSL SEMINAR - **SPECIAL TIME** - “Testing identity of distribution classes: Poisson Binomials””
http://illinois.edu/calendar/detail/3123/32092560
http://illinois.edu/calendar/detail/3123/32092560Mon, 29 Sep 2014 11:00:00 CDT**Abstract**
Distribution property testing has generated significant interest over the past decade. Given sample access to a distribution, the problem is to decide if the distribution has a property or is far from it. I will first give an overview of some distribution property testing problems such as testing for identity, uniformity, closeness of distributions.
The problem of testing identity of a single distribution has been studied extensively and was solved recently. However, very few results are known for the problem of testing if a distribution belongs to a class of distributions. In this work we consider the well-studied class of Poisson Binomial distributions, which are sums of $n$ independent, but not necessarily identical Bernoullis. We provide a sample near-optimal algorithm for testing whether a distribution supported on $\{0,...,n\}$ to which we have sample access is an unknown Poisson Binomial distribution, or far from all Poisson Binomial distributions. The sample complexity of our algorithm is $O(n^{1/4})$ to which we provide a matching information theoretic lower bound. Our algorithm improves quad ratically over the known results. This is one of the first optimal results on testing against a class of distributions.
Part of the talk will appear in the ACM-SIAM Symposium on Discrete Algorithms (SODA 2015), and is joint work with Constantinos Daskalakis, MIT.
******Bio******
Jayadev Acharya is a postdoctoral researcher with the theory of computation group in EECS, MIT. He obtained his Ph.D. in ECE from UC San Diego. He is interested in information theory and machine learning, particularly in designing sample optimal and time efficient algorithms for distribution learning and property testing. He is a recipient of the ISIT student paper award in 2010, and the Shannon Graduate Fellowship from UC San Diego in 2012.CSL SEMINAR - “Rateless Lossy Compression via the Extremes””
http://illinois.edu/calendar/detail/3123/32092558
http://illinois.edu/calendar/detail/3123/32092558Mon, 29 Sep 2014 16:00:00 CDT**Abstract**
I will begin by presenting a simple lossy compressor operating at near-zero rate: The encoder merely describes the indices of the few maximal source components, while the decoder’s reconstruction is a natural estimate of the source components based on this information. This scheme turns out to be near-optimal for the memoryless Gaussian source in the sense of achieving the zero-rate slope of its distortion-rate function.
Motivated by this finding, I will then introduce a scheme comprising of iterating the above lossy compressor on an appropriately transformed version of the difference between the source and its reconstruction from the previous iteration. The proposed scheme achieves the rate distortion function of the Gaussian memoryless source (under squared error distortion) when employed on any finite-variance ergodic source. Its storage and computation requirements are moderate at both the encoder and decoder. It further possesses desirable properties which I will discuss and we respectively refer to as infinitesimal successive refinability, ratelessness, and complete separability.
******Bio******
Albert No is a PhD candidate in the Department of Electrical Engineering at Stanford University, under the supervision of Prof. Tsachy Weissman. His research interests include relations between information and estimation theory, lossy compression, joint source-channel coding, and their applications. Albert received a Bachelor's degree in both Electrical Engineering and Mathematics from Seoul National University, in 2009, and a Masters degree in Electrical Engineering from Stanford University in 2012.