Publication and selected Projects:

Discovering RL agents in the Code Space: 

Project Overview:  Improving the efficacy of LLMs to create code that solves RL algorithms. We derive our inspiration from Quality Diversity Algorithms (MAP-elites) and Dreamcoder to enhance the effectiveness of Funsearch algorithm in searching through the code space. Done in collaboration with Prof. Eugene Vinitsky (NYU Tandon). 

Enhancing Context Through Contrast: 

Project Overview:  Proposed a novel post-pretraining step to enhance Neural Machine Translation performance. Based on the ideology that languages are transformation from an abstract meaning space. Done in collaboration with Kshitij Ambilduke (Université Paris-Saclay),  Rishika Bhagwatkar (Université de Montréal), Khurshed P. Fitter (EPFL), Prasad Vagdargi (JHU) and Prof. Sheetal Chiddarwar (VNIT). 

[Paper/Code]

Medical Visual Question Answering: 

Project Overview: Benchmarked  a number of  multimodal transformer architectures on ImageCLEF 2019 dataset using Meta AI Research's (then Facebook AI Research) Multi-Modal Framework (MMF). Done in collaboration with Prasad Vagdargi (final year PhD student, Johns Hopkins University) 

[Code]