Trapit Bansal
AI researcher focused on ML, NLP, and AI reasoning
San Francisco, California
Summary
Established academic researcher in machine learning and NLP with a strong publication record: his PhD work and associated papers focus on few-shot learning, self-supervised meta-learning, and NLP, and he has multiple conference publications (EMNLP, ACL, ICLR, NeurIPS). umass+2
Contributed to frontier reinforcement-learning and reasoning model work at OpenAI: listed among contributors and foundational participants on the o1 model series and associated system card, indicating involvement in building chain-of-thought / reasoning-focused models. arxiv+2
Experienced in both academic and industry research environments, with multiple research internships (Facebook, Google, OpenAI, Microsoft) during his graduate studies and a transition to industry research roles culminating in positions at OpenAI and Meta. trapitbansal+1
Recognized and well-cited researcher with significant collaborations: co-authors include senior figures in ML research (e.g., Andrew McCallum, Ilya Sutskever, Igor Mordatch) and a Google Scholar profile showing substantial citations. google+1
Work
Education
Projects
Writing
OpenAI o1 System Card
December 1, 2024System card describing the o1 model series: training with large-scale reinforcement learning for chain-of-thought reasoning, safety evaluations, red teaming, and preparedness considerations.
FEW-SHOT NATURAL LANGUAGE PROCESSING BY META-LEARNING (PhD dissertation)
January 1, 2022Doctoral thesis compiling contributions on few-shot NLP via meta-learning, including several conference papers and experiments.
Self-Supervised Meta-Learning for Few-Shot Natural Language Classification Tasks
January 1, 2020Describes methods for self-supervised meta-learning applied to few-shot NLP classification tasks; accepted to EMNLP (oral).
Continuous Adaptation via Meta-Learning in Nonstationary and Competitive Environments
January 1, 2017Presents a gradient-based meta-learning algorithm for adaptation in dynamically changing/adversarial multi-agent environments and introduces the RoboSumo environment for iterated adaptation games.