Syamantak Datta Gupta

 


 

Syamantak Datta Gupta

Postdoctoral Fellow

Computer Networks Research Laboratory (Room 702)

McConnell Engineering Building
McGill University
3480 University Street
Montréal, Québec
H3A 2A7 Canada

email: syamantak.dattagupta at mail.mcgill.ca

I am a Postdoctoral Fellow at the Department of Electrical and Computer Engineering, McGill University, under the co-supervision of Professors Mark Coates and Michael Rabbat.

I finished my PhD from the Department of Electrical and Computer Engineering, University of Waterloo under the supervision of Professor Ravi R. Mazumdar in January 2014. My PhD thesis was in the area of statistical signal processing and time series analysis.

I received my Master of Applied Science (MASc) from the Department of Electrical and Computer Engineering, University of Waterloo in 2009; and my Bachelor of Engineering (BE) degree from the Department of Electrical Engineering, Jadavpur University in 2006.

 

Research Interests:

  • Statistical signal processing
  • Particle filters and ensemble Kalman filters
  • Complex networks and Granger-causality graphs
  • Time series analysis and estimation
  • Information theory and directed information
  • Stochastic processes and probability theory

 

Research topics worked on:

Error propagation in gossip-based distributed particle filters

We examine the impact of the gossip procedure on distributed particle filters that employ an approximate version of the global likelihood function. For such algorithms, we derive time-uniform bounds on the estimation error and present associated exponential inequalities. These results allow us to evaluate the impact of such approximations on the overall performance of the distributed particle filter, and analyze its stability.

Inferring Granger-causality graphs among time series through pairwise tests

In this work, we consider interdependent groups of jointly wide sense stationary and cyclostationary discrete time stochastic processes and investigate the problem of quantifying their causal dependence relations using Granger-causality as a tool. It is shown that pairwise tests can provide useful and reasonably accurate information on the causal connections of the system at low computational costs.

A frequency domain Lasso approach for detecting interdependence relations among time series

An important consideration in determining causal dependence relations within a family of time series is to keep the resulting model sparse, as such a model is easy to interpret and more conducive to prediction. In this research, we derive a new technique that uses a frequency domain Lasso approach to find a linear estimator for a family of time series while preserving only the strongest dependence links and eliminating the weaker ones.

Asymptotic behaviour of the spectral density of finite order approximations of stationary time series

Approximation of a wide sense stationary time-series by finite order autoregressive (AR) and moving average (MA) sequences is a problem encountered in many applications. In this work, the asymptotic behavior of the spectral density of such approximations is studied. We consider two aspects: convergence of spectral density of moving average and autoregressive approximations when the covariances are known and when they are estimated. It is shown that under certain mild conditions on the spectral density and the covariance sequence, the spectral densities of both estimates converge as the order of approximation increases.

A comparative study of the particle filter and the ensemble Kalman filter

Non-linear Bayesian estimation, or estimation of the state of a non-linear stochastic system from a set of indirect noisy measurements is a problem encountered in several fields of science. The particle filter and the ensemble Kalman filter are both used to get sub-optimal solutions of Bayesian inference problems, particularly for high-dimensional non-Gaussian and non-linear models. In this work, it is shown that for the class of models considered, under certain assumptions, the two filters become methodologically analogous as the sample size goes to infinity.

Publications

Journal

  1. Syamantak Datta Gupta, Ravi R. Mazumdar and Peter W. Glynn, "On the Convergence of the Spectrum of Finite Order Approximations of Stationary Time Series", Journal of Multivariate Analysis (Volume-121, October 2013, pp-1-21).

Conference

  1. Syamantak Datta Gupta and Ravi R. Mazumdar "A frequency domain Lasso approach for Detecting Interdependence Relations among Time Series" , International work-conference on Time Series (ITISE) 2014.

  2. Syamantak Datta Gupta and Ravi R. Mazumdar "Inferring Granger-Causality Among Cyclostationary Time Series Through Time-invariant Estimators" , International work-conference on Time Series (ITISE) 2014.

  3. Syamantak Datta Gupta and Ravi R. Mazumdar "Inferring Causality in Networks of WSS Time Series by Pairwise Estimation Methods" (invited paper), In 2013 Information Theory and Applications Workshop (ITA 2013), University of San Diego, San Diego, February 10-14, 2013.

  4. Syamantak Datta Gupta and Ravi R. Mazumdar "On the convergence of the spectral density of autoregressive approximations via empirical covariance estimates", In 46th Annual Conference on Information Sciences and Systems (CISS' 2012), Princeton University, Princeton, March 21-23, 2012.

  5. Syamantak Datta Gupta, Ravi R. Mazumdar and Peter W. Glynn, "On the asymptotic behavior of the spectral density of autoregressive estimates", In 49th Annual Allerton Conference on Communication, Control and Computing, September 28-30, 2011.

  6. Syamantak Datta Gupta, Satadal Ghosh and Abhijit Sarkar. "Security of a highly protected zone against intruders: a novel data fusion model" International Conference on Advances in Control and Optimization of Dynamical Systems (ACODS' 2007), Indian Institute of Science, Bangalore, India, February 1-2, 2007.

Selected Talks

  1. "On the Convergence of Finite-Order Approximations of Stationary Time Series", At the Graduate Student Research Conference 2011, University of Waterloo, April 28, 2011.

 

Teaching Assistantships

Below is a summary of the courses I have been a teaching assistant to at the University of Waterloo. As a teaching assistant, my duties and responsibilities included guiding undergraduate students at the laboratory and evaluating them. The last column indicates the overall quality of my teaching, rated out of 5.0 by students.

Term Course Instructor Students Rating
Winter 2012 Communication Systems (ECE 318) Professor Patrick Mitran 221 4.47
Fall 2009 Linear Circuits (ECE 140) Professor Ramadan A. El-Shatshat 364 4.53
Spring 2009 Fundamentals of Electrical Engineering (ECE 100) Professor Ramadan A. El-Shatshat 274 4.16
Fall 2008 Sensors and Instrumentation (MTE 220) Professor James A. Barby 97 4.21
Spring 2008 Fundamentals of Electrical Engineering (ECE 100) Professor Ramadan A. El-Shatshat 237 4.31