Communication, Control and Signal Processing Seminar 


Abstracts of talks

Fall 2014

10/16 Katie Haymaker (Villanova University), Using a Graph Perspective to Investigate Bit Assignments for Codes

ABSTRACT: This talk will focus on applying (j,k)-regular LDPC codes to a memory setting with two different bit-error rates. We will discuss how changes in local connectivity to the two types of bits impacts decoding performance for a particular decoding algorithm. Using these results, we propose an assignment scheme that approaches the desired degree distributions. We will conclude by discussing variations of this problem that arise naturally in applications.

10/30, 11/6 Prakash Narayan (UMD), Interactive Function Computation

ABSTRACT: Information theoretic models for multiuser source and channel coding usually take the communication between multiple terminals to be ``simple" or autonomous. On the other hand, studies of multiparty function computation, especially in computer science, highlight the useful role of interactive communication. We shall describe basic structural properties of interactive communication. ``Single-shot" bounds will be presented for the amount of common randomness, i.e., shared information, that can be generated among the terminals using such communication. A few simple consequences with applications will be discussed. This talk is based on joint works with Imre Csiszár, Sirin Nitinawarat, Himanshu Tyagi and Shun Watanabe.

11/20 Marcos Vasconcelos (UMD), Distributed Estimation over the Collision Channel

ABSTRACT: Consider a system consisting of two sensors and a remote estimator connected by a network modeled by a collision channel. The collision channel is a simplified model for interference, which can only convey one packet per unit time and declares a collision when both sensors simultaneously transmit. In this talk, we will formulate a one-shot problem where the sensors measure independent random variables and must communicate them through the collision channel to a remote estimator, which is interested in forming an estimate of both measurements. We will show that the communication policies that minimize a mean squared error are deterministic threshold policies and that this structure is independent of the probability distribution of the random variables. We will then show how to explicitly compute the communication and estimation rules by using a modified version of the Lloyd-Max algorithm. If time permits, we will prove that this algorithm converges globally to a locally optimal solution when the random variables are Gaussian.
This is joint work with Prof. Nuno Martins.

12/04 Siddharth Pal (UMD), Learning in Games

ABSTRACT: We will start with a brief introduction to the language of game theory which will include discussions on strategic-form repeated games, various notions of equilibria and relevant convergence notions to such equilibria. Next, a brief overview of relevant learning rules will be given that are already available in the literature. The talk will move on to a description of a simple proposed learning rule in games. We demonstrate that this intuitive rule guarantees almost sure convergence to a pure-strategy Nash equilibrium in a large class of games we call generalized weakly acyclic games (GWAGs). We also show that the probability that the action profile does not converge to a pure-strategy Nash equilibrium decreases geometrically fast in the aforementioned class of games. The next half of the talk will look at a setting with delays in the game structure. We will look at various forms of delays and explore if the proposed learning rule can cope with such delays. Finally, if time permits, we will discuss various refinements of Nash equilibrium that is robust to “faulty” or “unexpected“ behavior and coalitions among agents. This talk is based on joint works with Prof. Richard La.

12/10 Sunav Choudhary (USC), Identifiability Analysis in Bilinear Inverse Problems with Application to Sparse Blind Deconvolution

ABSTRACT: A number of difficult non-linear inverse problems in signal processing, like blind deconvolution, matrix factorization, dictionary learning and blind source separation share the common characteristic of being bilinear inverse problems. A key concern for these inverse problems in applications like blind equalization in wireless communications and data mining in machine learning is that of identifiability of the generative model. In this talk, we exploit a connection to low-rank matrix recovery and develop a flexible and unifying framework for the analysis of identifiability in cone constrained bilinear inverse problems to derive sufficient conditions and scaling laws. For blind deconvolution in particular, our approach results in a measure theoretically tight partially parametric and partially recursive characterization of the ambiguity space leading to a surprising negative result on the identifiability of canonical-sparse blind deconvolution.
This is joint work with Prof. Urbashi Mitra at University of Southern California.

12/11 Biswadip Dey (UMD), Data Smoothing via Optimal Control

ABSTRACT: The problem of recovering continuous time signals from a set of discrete measurements is ill-posed in a classical sense (high sensitivity to noise, non-uniqueness of solution). However, this lack of well-posedness can be tackled through a regularized approach. The main idea behind regularization is to embed the problem of interest into a hypothesis space and minimize a cost functional expressed as a sum of two terms: (i) misfit of a hypothesis to observed data; and (ii) a penalty functional accounting for complexity of a hypothesis. In our context, we build the hypothesis space by introducing generative models governed by ordinary differential equations with inputs, states and outputs. This yields the hypothesis as output of the generative model, given an input (control), and we regularize this problem by trading total fit-error against suitable penalty functionals of input and state. This enables us to apply techniques from optimal control (e.g. Riccati equation, Pontryagin’s maximum principle) and obtain solutions in a (semi)-analytical way.
This talk is based on joint works with Prof. P. S. Krishnaprasad.