Communications Seminar


FALL 2010:
Time and Location: Thursdays 14:30-15:30, Location  Jeffery 115

  1. Thursday, October 28th, Maxim Raginsky (Duke University):

Title: Fundamental limitations of adaptive dynamical systems: an information-theoretic meta-framework.  
Abstract:  Adaptive dynamical systems arise in a multitude of contexts,e.g., optimization, control, communications, signal processing, andmachine learning. A precise characterization of their fundamental limitations is therefore of paramount importance. In this talk, I consider the general problem of adaptively controlling and/or identifying astochastic dynamical system, where a priori knowledge allows us to place the system in a subset of a metric space (the uncertainty set). I will present an information-theoretic meta-theorem that captures the trade-off between the metric complexity (or richness) of the uncertainty set, the amount of information acquired online in the process of controlling and observing the system, and the residual uncertainty remaining after the observations have been collected. Following the approach of G. Zames, I quantify a priori information by the Kolmogorov (metric) entropy of the uncertainty set, while the information acquired online is expressed as a sum of information divergences. I will then use the meta-theorem to derive new minimax lower bounds on the metric identification error, as well as to give a simple derivation of the minimum time needed to stabilize an uncertain stochastic linear system.

  1. Friday, November 5, Department Colloquium at 14:30, Aaron Wagner (Cornell University):

Title:  Estimation and Classification in the Learning-Limited Regime
Abstract: Motivated by models of natural language, we consider estimation and classification in the asymptotic regime where the source alphabet grows linearly with the number of observations. We show that, although the underlying distribution cannot be learned in this regime, it is possible to consistently estimate its entropy, the likelihood of the observed string, the divergence between the true and empirical distributions, and other similar quantities. We also show that consistent classification is possible, although standard tests such as the generalized likelihood ratio test, the chi-squared test, and the Hellinger test are all inconsistent in this regime. In fact, consistent classification is possible if the alphabet grows subquadratically with the number of observations, and it is impossible if the alphabet grows quadratically or faster.

This is joint work with Benjamin Kelly, Thitidej Tularak, Pramod Viswanath, and Sanjeev Kulkarni.

  1. Thursday, November 11,  

  2. Thursday, November 18, Yihong Wu (Princeton University):  

Title:  A Shannon Theory for Compressed Sensing
Abstract: Compressed sensing is an approach to lossless encoding of analog sources by real numbers rather than bits, dealing with efficient recovery of a sparse real vector from the information provided by linear measurements. Instead a worst-case (Hamming) approach, we develop a Shannon-theoretical framework for compressed sensing, where we model sources as random processes and adopt a source-coding (resp. joint source-channel coding) approach to noiseless (resp. noisy) compressed sensing problems. The information dimension and MMSE dimension of the source prove to be the fundamental limits of compression rate.

  1. Thursday, November 25, Johannes Karlsson ( Royal Institute of Technology (KTH)):

             Title: Low-delay sensing, transmission, and relaying
Abstract: We consider cooperative sensing and transmission in wireless sensor networks. Due to low-delay constraints, we turn to joint souce-channel  coding by means of memoryless optimized mappings. Design algorithms based on iterative optimization are proposed and evaluated empirically. Two cases are considered, namely, design for analog measurements and design for digital data. In both cases, we show that nonlinear, optimized mappings outperform linear uncoded transmission. By studying the structure of the optimized source–channel and relay mappings, we provide useful insights on how the optimized systems work.
  1. Thursday, December 2,