Habitat-Lab (Facebook Research)

Open Source Contribution

Area : Embodied AI

• Habitat-Lab is a modular high-level library to train embodied AI agents across a variety of tasks, environments, and simulators.
• Implemented a new task (Visual Language Navigation task) for the Room2Room dataset which allowed researchers to train embodied agents for this novel task on Habitat.​
• Implemented the DTW metric paper for Habitat which is used to measure the performance of embodied agents.​

Technologies Used: Python, Pytorch, pytest
Code

Exploration Based Top-k Join in PostgreSQL

Area : Databases

• Implemented an exploration based algorithm inside PostgreSQL to identify the distribution of the data.
• Depending on the distribution, a learning based approach like bandit join or nested loop can be used to return top-k join results.

Technologies Used: C, PostgreSQL
Code

Gym Join

Area : Databases, Reinforcement Learning

• Gym Environment for performing join between 2 tables with one being an agent and the other acting randomly.

Technologies Used: Python, OpenAi Gym
Code

Connectome Analysis

Area : Computational Biology

• Performed structural analysis on human connectome and chimpanzee connectome data.
• Implemented graphical measures such as centralities, motifs, clustering coefficients, max-flow, analysis of unique edges, etc. among others, and showcased differences between the two brain networks.

Technologies Used: Python, IGraph, NetworkX
Slides | Code