my-img
I am currently pursuing a PhD in Computer Science at McGill University and MILA under the supervision of Dr. Doina Precup.
Github / Google Scholar / LinkedIn / CV

Research Interests:

  • Multi Agent Reinforcement Learning
  • Model-based Reinforcement Learning
  • Relational Learning
  • Bayesian Modeling
I am currently working on coordination among agents in the multi-agent reinforcement learning setting. I am also interested in exploring representation learning and value-based planning methods. I previously worked on efficient approaches for modeling uncertainty and planning as part of my Master's thesis.

Prior to joining McGill, I worked as a Research Associate (2016-2018) at Indraprastha Institute of Information Technology, Delhi (IIIT-D) under the guidance of Professor A.V. Subramanyam. My work at IIIT-D revolved around devising general counter forensic methods for popular image processing operations such as Median Filtering. Furthermore, I briefly investigated few shot learning based approaches for the problem of Object Tracking.

I completed my Bachelor's in Instrumentation and Control Engineering (2011-2015) from Netaji Subhas Institute of Technology, Delhi University, following which I joined EXL Services (2015-2016) as a Business Analyst. My work at EXL involved validation of underwriting, balance consolidation and debt sale models used for generating financial insights as part of the credit card portfolio.

Publications:

Anti-forensics of Median Filtering and Contrast Enhancement, Journal of Visual Communication and Image Representation
  • The paper proposes an anti-forensic technique to counter spatial domain forensic detectors and demonstrates it's accuracy on popular image manipulation operations such as median filtering and contrast enhancement. The proposed optimization modifies the image so as to incorporate the median filtering or contrast enhancement operation while ensuring that the image's spatial characteristics do not change significantly.
Anti-forensic technique for median filtering using L1 - L2 TV model, IEEE International Workshop on Information Forensics and Security (WIFS 2016)
  • The paper proposes an anti-forensic technique for countering Median filtering forensics. The counter forensic attack is formulated as an optimization problem, constrained by the difference in Anisotropic and Isotropic Total Variation regularization. The optimization is written as difference of convex components and is solved using Split Bregman technique.