Projects

• Decentralized market making using Reinforcement Learning
  • Duration: August 2019 - December 2019
  • Affiliation: Equilibrium
  • Description: Successfully implemented pipe-lines and cleaned raw time-series data for feature generation (for eg, volume clocks on ETH/BTC tick data) to facilitate development of crypto-currency trading framework on decentralized exchanges and market making using deep reinforcement learning.
  • Links: Code
• Optimal area coverage using a swarm of robots in non-convex environments
  • Duration: May 2019 - August 2019
  • Affiliation: Dept. of Atomic Energy, Variable Energy Cyclotron Centre, Kolkata
  • Description: Worked at the intersection of control logic and swarm robotics. Studied and implemented various traditional approaches including Voronoi partitions, Lloyd’s algorithm, etc in a WeBots environment with e-Puck swarms using controllers built on Python and C. Successfully developed a decentralized algorithm leveraging concepts from these approaches coupled with a multi-agent scheme for enabling the swarm to jointly learn to cover a known non-convex area.
  • Links: Report
• Yield curve modeling and forecasting
  • Duration: April 2019 - August 2019
  • Affiliation: Axxela LLP
  • Description: Worked on forecasting the yield curve of US government treasury bonds. Various statistical frameworks (like PCA) and mathematical models (like Nelson-Siegel and Diebold-Li) were researched and innovative ways to capture information from the yield-curve data, like exploring chaos theory in financial markets and generation of images from rich time-series data were implemented, which were then leveraged by mathematical models ranging from stochastic models to neural networks. Implemented the Nelson-Siegel method for curve fitting on the yield curve of the US Treasury bonds. Parameters of the model were then forcasted using LSTMs.
  • Links: Code
• Autonomous Drone Navigation using Deep Reinforcement Learning
  • Duration: October 2018 - February 2019
  • Affiliation: EEE Department, BITS Pilani, Goa Campus
  • Description: Imitation Learning on IDSIA Dataset to classify directional commands for UAV to navigate. ResNet18 to classify images. Tested some features with Mask RCNN for segmentation of forest paths as navigable by UAV.
  • Links: Code