About Me
I will be joining Carnegie Mellon University (CMU) as a Master’s student pursuing MS in Intelligent Information Systems, in Fall 2023. I am highly interested in working on various research problems in the field of Machine Learning, Deep Learning, Natural Language Processing, Computer Vision, and Multimodal ML. Previously, I worked as an Associate Machine Learning Engineer at ExaWizards, where I worked on designing and deploying Natural Language query based Temporal Activity Localization systems. I graduated from IIIT-Hyderabad with B.Tech (Honours) + MS by Research in Computer Science. During my Master’s, I worked on Neural Machine Translation for Low-Resource languages and, Translation and Generation of Code-Mixed data, under the supervision of Professor Manish Shrivastava.
I have also worked with Professor Vineet Gandhi on improving the generalization ability of Image Classification models on unseen domains. Moreover, I worked on the detection of religious Hate Speech on online social media platforms such as Twitter, advised by Professor Ponnurangam Kumaraguru “PK” and Professor Jisun An.
News
- October 2022: Our work "Class-wise Domain Generalization: A Novel Framework for Evaluating Distributional Shift" accepted for publication at NeurIPS 2022 Workshop on Distribution Shifts.
- July 2022: Defended my Master's Thesis on "Neural Machine Translation for Low Resource Languages." [ Slides ] [ Blog ]
- December 2021: Our work "Reappraising Domain Generalization in Neural Networks" uploaded as a Pre-print.
- August 2021: Our work "A Dynamic Head Importance Computation Mechanism" accepted for publication at the Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021).
Experience
ExaWizardsAssociate Engineer, Machine Learning Working on Temporal Activity Localization in Videos. Specifically, the model takes a natural language description of the activity, and identifies the moments in the video where the activity happens. | |
Amazon IndiaApplied Scientist Intern Worked on improving the performance of Attribute Extraction models using Offline Reinforcement Learning (RL). Trained online RL agents to generate trajectories using its policy. Augmented the trajectories using the Trajectory Transformer model, to generate more representative trajectories. | |
PreCogResearch Assistant with Dr. Ponnurangam Kumaraguru & Dr. Jisun An Working on the detection of hate speech on online social media platforms such as Twitter. Specifically, we analyze the effects of offline events such as COVID-19 pandemic, on the spread of hateful content towards the Muslim community in India. | |
University of CalgaryResearch Intern with Dr. Hadi Hemmati Worked on Explainable Machine Learning. Designed and Implemented a method to generate explanations for sequence prediction tasks such as Code-Documentation generation. The explainable model was able to highlight the code-tokens and code-lines relevant to the generated documentation. | |
Sungkyunkwan UniversityResearch Intern with Dr. Hogun Park Worked on grounding knowledge graphs to impart commonsense knowledge to Question-Answering models for Multiple Choice QA task. We use Graph Convolution Networks to encode the information from the gounded knowledge graph, and use the dense representation along with the semantic featues for classification. The proposed new grounding method creates a dense schema graph, and enhances the performance of the QA models. [ GitHub ] | |
ExaWizardsAI Engineering Intern Worked on view-invariant body pose retrieval. Implemented a method to retrieve similar body poses of a given query pose. The method is able to identify poses with the same 3D pose but different 2D projections. | |
CVIT, IIIT-HyderabadResearch Assistant with Dr. Vineet Gandhi Working on improving the generalization ability of Image Classification models on unseen domains. Proposed a model that achieves state-of-the-art performance on multiple datasets. Also, proposed a new approach for evaluation that is closer to human judgment and more challenging than traditional methods. [ arxiv ] | |
LTRC, IIIT-HyderabadResearch Assistant with Dr. Manish Shrivastava Working on enhancing the performance of Neural Machine Translation models on low-resource languages. Proposed a novel approach for dynamically computing importance scores of various attention heads of the Transformer model [ Paper ]. I am also working on improving translation and generation of code-mixed data. |