- Working on all things, text like NER, closed domain multi-span question answering, summarization, multi-task learning, self-supervised learning, using few shot learning/meta learning for low resource tasks.
- Long standing collaboration with researchers from Columbia University and Google Brain working on projects to create fast, scalable attention mechanisms with provable convergence properties and applying them to diverse tasks like vision, audio, text and graphs. (ICLR-2022, ICML-2022, 2023)
- Previously worked on self-supervised learning on graph neural networks to predict diseases from single cell data (Accepted at ACM CHIL-20, Spotlight at ICML GRL+ 2020 workshop, AAAI '21)
- Applying DL to quantum mechanics and statistical physics problems (Accepted at AAAI-MLPS spring symposium 2021, 2 papers accepted at Neurips ML4Physical Sciences)
- Worked on various generative models like Normalizing Flows, VAEs and GANs.
- Created GLYPH + BERT by augmenting BERT by encoded images of Chinese characters which improved NER performance on Chinese Onto notes by +1 and on Weibo by +3 over the baseline BERT model at Human Language Technology Center of Excellence, JHU. (Accepted at AAAI-20)
- Graduated with a PhD in mathematics with a passion for teaching, learning and research. My thesis is focused on modular Galois deformations and Iwasawa theory.
- Also interested in using mathematical ideas to tackle gerrymandering and other social justice issues.