Current Projects

New York University Incremental Skill Discovery for RL Agents
I’m working in the CILVR Lab with Prof. Lerrel Pinto on expanding Citation:  et. al., & (). One After Another: Learning Incremental Skills for a Changing World. arXiv preprint arXiv:2203.11176. ’s incremental skill discovery framework to manipulation environments like AllegroHand.

New York University Speech Recognition for Mandarin Conversations
I’m working with Prof. Michael Picheny on DARPA’s CCU Program to improve speech recognition systems for conversational Mandarin, primarily on the AISHELL4 dataset (Citation:  et. al., , , , , , , , , , & (). AISHELL-4: An Open Source Dataset for Speech Enhancement, Separation, Recognition and Speaker Diarization in Conference Scenario. arXiv preprint arXiv:2104.03603. ).

Past Projects

Flixstock India Bust Replacement on Photorealistic Model Images
I worked with StyleGANs to generate bust-level crops of photorealistic model images, making use of techniques like GAN inversion and style mixing. I also used classical methods like MLS (Citation:  et. al., , & (). Image Deformation Using Moving Least Squares. In ACM SIGGRAPH 2006 Papers. (pp. 533–540). ) to create an end-to-end pipeline for bust replacement upto 1024px.

IBM Research Neural Conversational Question Answering
I worked with Danish Contractor on improving question answering systems on the ShARC dataset (Citation:  et. al., , , , , , , & (). Interpretation of Natural Language Rules in Conversational Machine Reading. arXiv preprint arXiv:1809.01494. ). This led to a patent (US 16/892805), and a paper (Citation:  et. al., , , , , & (). Neural Conversational QA: Learning to Reason vs Exploiting Patterns. Association for Computational Linguistics. ).

Even Further

Microsoft Research Code to Natural Language
[Report] I worked with Navin Goyal on techniques to describe code (SQL Queries and bash commands in particular) in natural language as part of my undergrad thesis.

IBM Research Question Generation
[Report] I worked on improving techniques to generate answerable questions from text.