-
Multihop Question Answering
Reproduced experiments for two published papers, DecompRC and Graph-based Recurrent Retriever and verified their validity and conclusions for the task of multihop question answering. Experimented by combining the merits of the two models to develop and evaluate an alternative approach to solving the same task and presented our results on the HotpotQA dataset.
-
Grayscale Video Colorization
Built a system that automatically colorizes grayscale videos and historical films, in an effort to restore videos to their original colors.
-
Deep Reinforcement Learning to Mastermind Settlers of Catan
Developed an agent to play the game, Settlers of Catan in its entirety, using a Deep Q-Network agent that can utilize gameplay mechanics and outperform both current state-of-the-art agents and humans.
-
Visual Relationship Detection
Proposed an approach to detect objects of interest in an image, given a specific predicate relationship that has been highlighted through text input.
-
Visual Essence
Working on vision research in the creative AI space under Dr. Devi Parikh, focusing on obtaining a visual representation that adequately conveys the essence of a person’s description of themselves.
-
AlphaZero AI to play Blind Multi-Chess
Developed an AI agent to play the game of Reconnaissance Blind Multi-Chess (RBMC) by learning to play against Stockfish.
-
Information Retrieval across Indic Scripts
Developed an Indic script web search engine (Hindi, Tamil, English) that involved querying and retrieving results in more than one language. Proposed a new ranking algorithm based on query intent by applying classification and heuristics on multilingual queries. This project was submitted and accepted as my undergraduate thesis under the advise of Dr. Ranjani Parthasarathi.