Rachit Bansal
Kempner Graduate Fellow
PhD Student in Computer Science
Contact Information
Areas I Research:
About
Rachit Bansal is a PhD student in the ML Foundations group, advised by David Alvarez-Melis, Martin Wattenberg, and Fernanda Viegas. Prior to joining Harvard, he was a pre-doctoral researcher at Google DeepMind, working on modular training of large models.
Bansal did his undergraduate studies at Delhi Technological University in New Delhi, India. During this time, he was fortunate to have pursued a thesis at the Technion, Israel, with Prof. Yonatan Belinkov. He was also a part of a couple of wonderful collaborations with Naomi Saphra (NYU) and Danish Pruthi (CMU).
Research Focus
Bansal’s research involves the study of mechanisms to enhance understanding and usability of current day large-scale models.
On the understanding end, he is deeply interested in understanding the learning dynamics on neural networks, with a particular focus on generalization. On the usability side, he is interested in building modular (and non-monolithic) language models that are lightweight and easy to adapt for downstream use-cases (knowledge, domains). He is excited about furthering modularity to improve efficiency and control of current models.