Research Interests 


I am interested in understanding the theoretical and practical aspects of information processing and decision making in human-machine systems with potentially unreliable components. Tools from statistical signal processing, information theory, error correcting codes, and statistical learning, are used in my research. Due to the nature of my research, I have also been actively involved in interdisciplinary collaboration with researchers from other fields, such as psychology.

Research Projects


Human-Machine Inference Networks
  • Research on design and analysis of reliable human-machine inference systems consisting of humans and machines which complement each others' strengths.
  • Presence of unskilled humans and/or malicious machines can deteriorate the system performance drastically.
  • By understanding the behavior of humans in such systems, the machine parameters can be optimized for improved performance.
The Non-Regular CEO Problem
  • Research on the non-regular CEO problem where the CEO could be interested in estimating the belief that a particular event takes place, for which she uses multiple subordinates to send their observations.
  • By modeling the noisy versions using copula, a $1/R^2$ convergence of the distortion is established, an intermediate regime between the exponential convergence of discrete case and $1/R$ convergence of Gaussian case.
  • Achievability is proved by a layered architecture with quantization, entropy coding, and midrange estimator, and converse is proved using the Chazan-Zakai-Ziv bound.
Reliable Crowdsourcing
  • Designed crowdsourcing systems to ensure reliable classification despite unreliable crowd workers.
  • Coding theory based techniques are used to pose easy-to-answer binary questions to the crowd workers.
  • Three different crowdsourcing models are considered: systems with independent crowd workers, systems with peer-dependent reward schemes, and systems where workers have common sources of information.
  • Showed that pairing among workers and diversification of the questions help in improving system performance.
Distributed Inference with Byzantine Data
  • Research on the effect of and mitigation of malicious sensors (Byzantines) on Distributed Inference Networks.
  • Interactions between the honest and the malicious sensors are modeled using game theory and optimal attack strategies are determined to characterize the effect of such Byzantines in the network.
  • Machine learning and statistical signal processing schemes are used to mitigate the effects of Byzantines by learning their behavior and identifying the malicious sensors.
Coding Theory for Reliable Signal Processing
  • Research on design and analysis of systems to ensure reliable signal processing using coding theoretic ideas.
  • Effect of unreliable components in the network can be abstracted as errors in data.
  • Error-correcting based schemes can be designed to correct these errors due and ensure reliable performance.
Decision Fusion by Humans
  • Developed experiments to understand the phenomenon of decision fusion by humans.
  • Compare the behavior of humans and sensors/machines while performing this task of decision fusion.
  • Observed that a deterministic rule used by machines does not characterize the human behavior, which is not deterministic in nature.
  • Developing randomized decision rule where the decision is first determined by a deterministic rule which then goes through a binary symmetric channel.
Quantizer Design for Distributed Estimation
  • Explored the design of optimal quantizers in distributed estimation under the Bayesian criterion.
  • For a conditionally unbiased and efficient estimator at the fusion center, it is optimal to partition the local sensors into groups, with all sensors within a group using the same quantization rule.
  • For capacity constrained wireless network, binary quantizers at the local sensors are optimal under certain conditions.
  • Also derived the optimality conditions of quantizers for conditionally dependent observations.
Distributed Inference in Tree Networks
  • Developed simple-to-implement coding theory based techniques to solve the distributed inference (classification and estimation) problems in tree structures.
  • Studied the asymptotic inference performance of the proposed schemes for two different classes of tree networks: fixed height tree networks, and fixed degree tree networks.
  • Showed that the proposed schemes are asymptotically optimal under certain conditions.
Sensor Selection in Field Reconstruction
  • Introduced a new sparsity-promoting penalty function for sensor selection problems in field reconstruction.
  • Using a reweighted $\ell_1$ relaxation of the $\ell_0$ norm, the sensor selection problem is reformulated as a convex quadratic program.
  • Presented two fast algorithms: accelerated proximal gradient method and alternating direction method of multipliers, in order to handle large-scale problems.