Projects

GPGPU solutions for Linear Least Squares Problem

CSE 6643 - Numerical Linear Algebra, Spring 2018
Guide : Prof. Haesun Park

  • Implemented three general purpose GPU solutions for the linear least squares problem — Householder QR decomposition, Cholesky decomposition and Givens QR decomposition —  and their CPU counterparts for comparison on a 2D pose graph optimization problem solvable by Newton's method

  • Report and Presentation

Kernel Dictionary Learning

Bachelor's Thesis, Fall 2016 - Spring 2017
Guide : Prof. Suyash P. Awate

  • Implemented a kernel dictionary learning algorithm for data on spherical manifolds

  • Demonstrated effective application in image denoising and image classification tasks

  • Studied the effect of different regularizers and kernels, on robustness in classification performance of the algorithm, under different kinds and intensities of noise, on MNIST handwritten digits dataset

  • Thesis Report and Presentation

Medical Image Segmentation : DeepCut

CS 736 - Algorithms for Medical Image Processing, Spring 2017
Guide : Prof. Suyash P. Awate

  • Implemented DeepCut segmentation algorithm for finding segmentation of the heart from human chest MR images, using user-input bounding box annotations

  • Used an iterative procedure of fuzzy pixel mask generation using a conv net and refinement using a dense conditional random field (CRF

  • Report

Reinforcement Learning : Carrom playing bot

CS 747 - Foundations of Intelligent and Learning Agents, Fall 2016
Guide : Prof. Shivaram Kalyanakrishnan

  • Implemented and evaluated three approaches of building a carrom playing bot  —  deep Q-learning, deep deterministic policy gradients and using hand-coded heuristics

  • Report

Equillibria in multiplayer timed games

RnD project, Fall 2015
Guide : Prof. Krishna S.

  • Studied different equilibria in sequential non-competitive multiplayer games on timed automata

  • Considering only memoryless player strategies, proved undecidability of the existence of a cost bounded Nash, Stackelberg or Incentive equilibrium in a 2 player sequential timed game with 3 clocks (a result that trivially extends to more players or clocks)

  • Report