Shashwat Rai

Experience

• 

  The Institute of Mathematical Sciences, Chennai

  Jun 2022 - Feb 2023

  Research Intern

  I was an intern at ASlab, a Max Planck Research Group at the Institute of Mathematical Sciences. I was working on boolean logic network modelling and computational biology. I further explored Petri nets, differential equations and other advanced mathematical models. I was also able to learn and explore further research topics in theoretical computer science and mathematics. I also was exposed to graph Ricci curvatures, discrete morse theory and statistical physics. I also participated in workshops and seminars on theoretical computer science and mathematics Show less

• 

  Dana-Farber Cancer Institute

  Sept 2023 - Mar 2024

  Undergraduate Researcher

  I am a research assistant at the Hormoz Lab of the Department of data science at the Dana Farber Cancer Institute. Where I am working on mathematical and statistical methods to extrapolate causation from noisy biological data. I am also working on machine learning approaches to interpret fluctuations in large scale data models.

• 

  Harvard John A. Paulson School of Engineering and Applied Sciences

  Jan 2024 - now

  Undergraduate Researcher

  This project aims to investigate how AI systems can adapt to new, shifting data through continual learning, pretraining, and fine-tuning. It will explore how foundation models in AI and robotics can be robust against distribution shifts, a core challenge in machine learning, where test data often diverge from training data. The research will analyze strategies for AI to better generalize and maintain performance in dynamic environments, contributing to the development of more flexible systems. The primary aim for the project would be to analyze and test various parameter efficient fine-tuning algorithms using the WILDS dataset distribution shifts. We would also test current finetuning algorithms like SGD and Adams see whether catastrophic memory loss arises when dataset distribution shifts occur. Further, we would use theoretical methods that have not yet been explored by roboticists to solve the problems that arise with real world fine tuning in Robotic learning and continual learning environments . Show less

• 

  Harvard Economics Department

  Jan 2024 - Jun 2024

  Research Assistant

  This project uses a large language model, trained to capture semantic similarity, to measure the spread of economic ideas. Using an extensive corpus of economics journal articles, book reviews, and dissertations stretching back to the early 20th century, we use semantics to measure which ideas have been influential in economics, validating this approach with citation networks. We then track the spread of economic ideas through tens of millions of pages of U.S. government documents (to measure how ideas spread from academia to policy), U.S. local newspapers (to measure popular influence), corporate documents (to measure how ideas spread to the private sector), and writings of other social scientists (to measure cross-disciplinary fertilization). Our initial application is to the United States, but time permitting we also plan to begin studying how economic ideas over the past 100 years have influenced thinking on development policy in lower income countries. This is a part of the Semester Undergraduate Program in Economics Research and was selected through a competitive process. Show less

• 

  Kempner Institute at Harvard University

  Aug 2024 - now

  Undergraduate Researcher

  - Building a computational model of pragmatics and linguistics to interpret and the evaluate the ability of large language models (LLMs) to capture human-like eliciture patterns, which would contribute a novel evaluation of LLMs’ linguistic and common sense reasoning abilities.- Developing a probabilistic model of referring expression generation in order to explain the human behavioral data, drawing upon Bayesian models of communication such as the Rational Speech Act framework.

• 

  Harvard Radcliffe Institute

  Aug 2024 - now

  Radcliffe Research Partner

  -working on methods to leverage artificial intelligence to enhance fairness and efficiency within the justice system- working on using statistical methods and machine learning to understand legal data and trends -Selected through a competitive process to be a Radcliffe Research partner