machine learning

I am interested in using Machine Learning for social good. More specifically I like doing research in Natural Language Processing and Computer Vision. I’m also interested in exploring questions surrounding the ethics of, and accesibility in AI in more depth.

Machine Learning assisted evolutionary optimization (with Prof. Dhish Kumar Saxena , IIT Roorkee)

Multi-objective optimization involves optimizing multiple conflicting objectives simultaneously. Minimizing cost and fuel consumption, and maximizing user comfort of a vehicle is an example of such problems. Evolutionary multiobjective optimization algorithms effectively handle such problems, by iteratively aiming to achieve a well-converged and well-diverse set of Pareto-optimal solutions. Much research has been performed on the former. We analyse the suitability of machine learning algorithms to generate new solutions that are potentially more converged and/or more diverse. \\ In this project, we aim to investigate the dataset features and choose a suitable ML model for this task, followed by an experimental investigation. We aim to isolate the parameters used by these optimizers and design a suitable adaptation function to adapt them on the fly, removing any prior fixations. Our work was accepted in EMO 2023, held in Leiden, Netherlands.

FaIRCoP: Facial Image Retrieval using Contrastive Personalization(with Prof. Rajiv Ratn Shah ,MIDAS Labs, IIIT Delhi)

Worked on Facial Image Retrieval for suspect identification. We proposed a relevance feedback method that utlizes the user's binary feedback to exploit the contrastive learning paradigm for encapsulating each user's personalized notion of similarity. For this, we worked on a novel loss function, optimized online via user feedback and a corresponding scoring function for inference. We also design a testbed to utilize simulated user feedback and automate the comparison of different relevance feedback mechanisms and other design choices.

Our work was accepted in the student abstract track of AAAI 2022 and WACV 2023

Making Discriminative Natural Language Infilling more robust(with Prof. Soujanya Poria ,Declare Lab, Singapore University of Technology and Design)

The main task is to generate semantically similar correct and incorrect options for masked sentences within common-sense stories and natural dialogue datasets and to make it as hard as possible for state-of-the-art NLP models to differentiate between the correct and incorrect options

Covid cases predictor and NPI prescriptor(with Prof. Debasish Ghose ,IISc Bangalore)

Built a Covid cases predictor using LSTMs and an NPI prescriptor model using an encoder-decoder architecture. Ranked among 48 finalists from 17 countries in the Global Pandemic Response Challenge, was a member of India's only qualifying team.