doi.bio/vedant_misra

Vedant Misra

Overview

Vedant Misra is a researcher, engineer, and entrepreneur with a background in developing and shipping high-value automated systems by leveraging and advancing deep learning. Misra is currently a researcher at Google, focusing on artificial intelligence and studying and extending the capabilities of large neural networks.

Education

Misra attended Columbia University, where he studied physics, mathematics, and economics as a Rabi Scholar. During his time at Columbia, he also worked on black hole research at the Institute for Strings, Cosmology, and Astroparticle Physics and studied complex systems at the independent think tank NECSI.

Career

Prior to his role at Google, Misra held various positions in the field of artificial intelligence and machine learning. He co-founded and served as CEO of Kemvi, a company focused on using machine learning to understand and personalize language at scale, which was acquired by HubSpot. At HubSpot, he led the Machine Learning Labs team, developing new products with text, images, and audio.

Misra also worked at OpenAI, overseeing the Algorithms and Reasoning teams and collaborating with Ilya Sutskever to lead the Reasoning, Algorithms, and Multimodal teams. His research interests include generative multimodal models, theoretical astrophysics, statistical finance, human-computer interaction, and network science.

Publications and Projects

Misra has authored and contributed to several publications and projects in the field of artificial intelligence and machine learning. Some of his notable works include:

"Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models" - a benchmark to evaluate and direct future research on language models, addressing socially harmful effects and disruptive capabilities.
"PaLM: Scaling Language Modeling with Pathways" - a study on the impact of scale on few-shot learning, introducing the Pathways Language Model (PaLM) and discussing ethical considerations.
"Exploring Length Generalization in Large Language Models" - an investigation into the length generalization capabilities of transformer-based language models and the combination of pre-trained models with scratchpad prompting.
"Solving Quantitative Reasoning Problems with Language Models" - the introduction of Minerva, a large language model pre-trained on general natural language data and further trained on technical content, to improve performance on quantitative reasoning tasks.
DALL-E - a system for generating images from text, developed by Aditya Ramesh.
Grokking - a study on the generalization of neural networks on small algorithmically generated datasets, demonstrating the process of "grokking" a pattern and improving generalization performance.

Conclusion

Vedant Misra is a prominent researcher and entrepreneur in the field of artificial intelligence and machine learning. With a strong educational background and a diverse range of experiences, he has contributed to advancing the state-of-the-art in deep learning and understanding the computational principles of intelligence.