PeopleofSuperintelligence

Researcher

Aaron Defazio

Aaron Defazio is a machine-learning researcher whose optimization methods reduce learning-rate tuning and training-schedule choices for AI. At the Australian National University, he completed an honours thesis on network topology in 2010, published a paper on scale-free networks in 2012, and introduced Finito and SAGA in his 2014 PhD thesis. He coauthored MADGRAD in 2022 and Schedule-Free in 2024; D-Adaptation won the 2023 ICML Outstanding Paper Award. At the University of Georgia's AI Research Day in 2025, he gave a keynote as Senior Staff AI Research Scientist at Meta. He published ScheduleFree+ in May 2026 and is a Member of Technical Staff at OpenAI as of October 2026.

Now at
OpenAI
Badges

Timeline

  1. 2010

    Network Topology Tomography

    At the Australian National University, he completes the honours thesis Network Topology Tomography in 2010 under Tiberio Caetano's supervision.

  2. 2012

    Learning scale-free networks

    With Tiberio Caetano, he presents a convex formulation for learning scale-free networks in a 2012 NIPS paper.

  3. 2014

    A PhD thesis on optimization

    He completes a PhD at the Australian National University in 2014 under Tiberio Caetano; his thesis introduces Finito and SAGA for machine-learning optimization.

  4. 2022

    The MADGRAD optimizer

    With Samy Jelassi, he introduces MADGRAD, an AdaGrad-family method for adaptive stochastic optimization, in 2022.

  5. 2023

    D-Adaptation wins an ICML award

    With Konstantin Mishchenko, he presents D-Adaptation, a single-loop method that sets the learning rate without knowing the distance from the initial point to the solution set. The work receives the 2023 ICML Outstanding Paper Award.

  6. 2024

    Schedule-Free learning

    With five coauthors, he proposes Schedule-Free in 2024, a method that avoids choosing a training stop step and adds no hyperparameters beyond standard optimizers with momentum.

  7. 2025

    Theory to Practice at FAIR

    At the University of Georgia's AI Research Day in 2025, he gives a keynote titled ‘Theory to Practice at FAIR: AI training algorithms’ as Senior Staff AI Research Scientist at Meta.

  8. As of October 2026

    OpenAI and ScheduleFree+

    In May 2026, he introduces ScheduleFree+, a learning-rate-free and schedule-free method for training large language models. As of October 2026, he is a Member of Technical Staff at OpenAI.

Career

  1. NowOpenAIMember of Technical Staff
  2. From 2025MetaSenior Staff AI Research Scientist

Education

  1. 2014Australian National UniversityPhD
  2. 2010Australian National UniversityHonours

From the biography.

  1. 22 of 133 people shaping AI with a former OpenAI role now work at AnthropicIncludes Aaron Defazio. Our estimates of where OpenAI recruits came from and where 133 people with former OpenAI roles work now, including the companies and labs they founded.
  2. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Aaron Defazio. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  3. 16% of AI career moves in the 2020s went from a university to a companyIncludes Aaron Defazio. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  4. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Aaron Defazio. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Aaron Defazio yet. Send us a free-licensed photo.

  1. Publications | Aaron Defazio aarondefazio.com. Checked 21 October 2026.
  2. A Momentumized, Adaptive, Dual Averaged Gradient Method jmlr.org. Checked 21 October 2026.
  3. ICML 2023 Awards icml.cc. Checked 21 October 2026.
  4. The Road Less Scheduled proceedings.neurips.cc. Checked 21 October 2026.
  5. AI Research Day 2025 | Artificial Intelligence ai.uga.edu. Checked 21 October 2026.
  6. ScheduleFree+: Scaling Learning-Rate-Free & Schedule-Free Learning to Large Language Models arxiv.org. Checked 21 October 2026.
  7. Aaron Defazio, PhD Scientist aarondefazio.com. Checked 21 October 2026.