Mikhail Yurochkin

Mikhail Yurochkin is a Staff AI Scientist at the IFM MBZUAI Silicon Valley Lab, where he leads data mixing for LLM pre-training. Previously, he served as Research Manager at the MIT-IBM Watson AI Lab, leading the Statistical Large Language Modeling group. He holds a PhD in Statistics from the University of Michigan, advised by Prof. Long Nguyen.

Education:

• PhD in Statistics, University of Michigan (advisor: Prof. Long Nguyen)

Research Interests: His work focuses on foundational and applied aspects of large language models (LLMs), including pre/post-training methodologies, data quality, reasoning, evaluation, routing, and efficient inference. He also explores statistical modeling approaches in areas like OOD generalization, algorithmic fairness, federated learning, and Bayesian nonparametrics. His contributions often bridge theoretical advancements with practical implementations in AI systems.

Key Contributions:

• Developed frameworks for LLM routing and compression (e.g., CARROT, LLM routing via optimal transport)
• Pioneered tinyBenchmarks for efficient LLM evaluation
• Advanced techniques for fairness in machine learning (e.g., individual fairness, distribution shifts)
• Contributed to federated learning through models like Federated Learning: A Comprehensive Overview

Labs & Affiliations: Active in collaborative environments such as the MIT-IBM Watson AI Lab and MBZUAI’s Silicon Valley Lab, focusing on cutting-edge research in AI alignment, scalable models, and ethical AI systems.