About

I am a Senior Research Scientist at the Institute of Foundation Models (MBZUAI) in Sunnyvale, California, USA, where I work on the data team building pipelines that generate synthetic data for large language model training. I received my Ph.D. in Statistics from the University of Michigan in 2026, advised by Yuekai Sun and Moulinath Banerjee, where I built efficient, statistically principled methods for evaluating AI systems and large language models — measuring what they can and cannot do.

My work draws on psychometrics, latent variable models, and statistical inference, and has appeared at ICML, NeurIPS, ICLR, and AISTATS, with more forthcoming in the Annals of Applied Statistics (Google Scholar). I have interned at Google DeepMind and the MIT-IBM Watson AI Lab, and my dissertation was supported by the Rackham Predoctoral Fellowship, awarded to University of Michigan doctoral students whose dissertations are unusually creative, ambitious, and impactful.

Feel free to reach out at maiapolo [at] umich [dot] edu.

Interests

  • AI Evaluation
  • Synthetic Data for LLMs
  • LLM-as-a-Judge
  • Psychometrics
  • Latent Variable Models
  • Statistical Inference
  • Machine Learning

Education

🎓 Ph.D. in Statistics
University of Michigan · 2021–2026
Advisors: Yuekai Sun & Moulinath Banerjee
🎓 M.Sc. in Statistics
University of São Paulo · 2021
Advisor: Renato Vicente
🎓 B.Sc. in Economics
University of São Paulo · 2018 · Best Student Award (ranked 1st of 49)
Advisor: Daniel Domingues dos Santos
🌏 Exchange Program
University of Tokyo · 2017–18 · merit scholarship

News & Events

Experience

2026
Senior Research Scientist — Institute of Foundation Models (MBZUAI)
Sunnyvale, California, USA · Sep 2026–present
Data team; building pipelines that generate synthetic data for large language model training.
2025
Student Researcher — Google DeepMind
London, UK · Aug–Dec 2025
Efficient AI model evaluation with Isabela Albuquerque; developed a statistical framework cutting human-annotation cost by 10–20× while preserving fine-grained capability measurement.
2025
AI Research Summer Intern — MIT-IBM Watson AI Lab
Cambridge, USA · May–Aug 2025
Game-based LLM evaluation with Leshem Choshen and Kristjan Greenewald; inferred model skills from dynamic game data (e.g., chess).
2024
AI Research Summer Intern — MIT-IBM Watson AI Lab
Cambridge, USA · May–Aug 2024
Scaling laws for LLM skills with Mikhail Yurochkin, predicting multi-benchmark performance across model families (Sloth).
2023
Research Assistant — University of Michigan
Ann Arbor, USA · 2023–2026
Developing new methods in statistics and machine learning with Yuekai Sun and Moulinath Banerjee.
2020
Research Assistant — Advanced Institute for Artificial Intelligence (AI2)
São Paulo, Brazil · 2020–2021
Credit-scoring research with the LatAm Experian DataLab data science team.

Selected Publications

2026
A Latent Variable Framework for Scaling Laws in Large Language Models
P. Cai, C. Cui, F. Maia Polo, S. Somerstep, L. Choshen, M. Yurochkin, M. Banerjee, Y. Sun, K. M. Tan, G. Xu
Annals of Applied Statistics · accepted, to appear
2026
CARROT: A Cost-Aware Rate-Optimal Router
S. Somerstep, F. Maia Polo, A. F. M. de Oliveira, P. Mangal, M. Silva, O. Bhardwaj, M. Yurochkin, S. Maity
Annals of Applied Statistics · accepted, to appear
2026
Rich Insights from Cheap Signals: Efficient Evaluations via Tensor Factorization
F. Maia Polo, A. Nematzadeh, V. Aglietti, A. Fisch, I. Albuquerque
Neural Information Processing Systems (NeurIPS) 2026 · accepted, to appear
2025
Bridging Human and LLM Judgments: Understanding and Narrowing the Gap
F. Maia Polo, X. Wang, M. Yurochkin, G. Xu, M. Banerjee, Y. Sun
Neural Information Processing Systems (NeurIPS) 2025
2025
Sloth: Scaling Laws for LLM Skills to Predict Multi-Benchmark Performance Across Families
F. Maia Polo, S. Somerstep, L. Choshen, Y. Sun, M. Yurochkin
Neural Information Processing Systems (NeurIPS) 2025
2025
A Statistical Framework for Weak-to-Strong Generalization
S. Somerstep, F. Maia Polo, M. Banerjee, Y. Ritov, M. Yurochkin, Y. Sun
International Conference on Learning Representations (ICLR) 2025
2024
tinyBenchmarks: Evaluating LLMs with Fewer Examples
F. Maia Polo, L. Weber, L. Choshen, Y. Sun, G. Xu, M. Yurochkin
International Conference on Machine Learning (ICML) 2024
2024
Efficient Multi-Prompt Evaluation of LLMs
F. Maia Polo, R. Xu, L. Weber, M. Silva, O. Bhardwaj, L. Choshen, et al.
Neural Information Processing Systems (NeurIPS) 2024
2024
Fusing Models with Complementary Expertise
H. Wang, F. Maia Polo, Y. Sun, S. Kundu, E. Xing, M. Yurochkin
International Conference on Learning Representations (ICLR) 2024
2023
Conditional Independence Testing under Misspecified Inductive Biases · spotlight
F. Maia Polo, Y. Sun, M. Banerjee
Neural Information Processing Systems (NeurIPS) 2023
See all publications on Google Scholar →

Awards

Contact

Email: maiapolo [at] umich [dot] edu

Institute of Foundation Models (MBZUAI), Sunnyvale, California, USA