PeopleofSuperintelligence

Researcher

Aryo Lotfi

Also known as Aryo Lotfi Jandaghi

Aryo Lotfi is a machine-learning researcher whose work studies reasoning in neural networks and large language models. He competed in math Olympiads, including the International Mathematical Olympiad, studied computer engineering at Sharif, and coauthored a 2022 NeurIPS paper on how neural networks learn logical functions. In 2023, his coauthored paper Generalization on the Unseen, Logic Reasoning and Degree Curriculum won ICML's Outstanding Paper Award, and EPFL awarded him an Apple AI/ML PhD Fellowship in 2024. At EPFL, he completed a computer science PhD under Emmanuel Abbe in 2025; his research spans Chain-of-Sketch, Transformer reasoning, adaptive reinforcement learning, and training curricula for language models. As of October 2026, he is a Research Scientist at Apple Machine Learning Research.

Now at
Apple
Badges

Timeline

  1. 2022

    Learning to reason with networks

    He coauthors Learning to Reason with Neural Networks: Generalization, Unseen Data and Boolean Measures, a NeurIPS 2022 paper that studies how gradient descent trains neural networks to learn logical functions and generalize under distribution shift.

  2. 2023

    An ICML Outstanding Paper

    At ICML 2023, he and Emmanuel Abbe, Samy Bengio, and Kevin Rizk receive the Outstanding Paper Award for Generalization on the Unseen, Logic Reasoning and Degree Curriculum.

  3. 2024

    Apple AI/ML fellowship

    In 2024, he receives the Apple AI/ML PhD Fellowship in ML Theory as a PhD student at EPFL's Chair of Mathematical Data Science.

  4. 2024

    Scratchpads for global reasoning

    In 2024, he coauthors Chain-of-Sketch: Enabling Global Visual Reasoning, which breaks global vision tasks into intermediate steps and reports improved out-of-distribution generalization with its inductive method. At NeurIPS 2024, he also coauthors How Far Can Transformers Reason? The Globality Barrier and Inductive Scratchpad, which shows that Transformers struggle to compose long chains of syllogisms.

  5. 2025

    RL for reasoning with rationales

    In 2025, he coauthors RL for Reasoning by Adaptively Revealing Rationales, a reinforcement-learning method that reveals correct partial solutions and adjusts their length using the model's reward signal.

  6. August 2025

    A PhD thesis on reasoning

    In August 2025, he defends Learning to Reason with Neural Networks: Principles and Methods in EPFL's Computer and Communication Sciences doctoral program under Emmanuel Abbe. Before his PhD, he studies computer engineering at Sharif University of Technology in Iran and competes in math Olympiads, including the International Mathematical Olympiad, during high school.

  7. 2026

    Goldilocks RL for reasoning

    In 2026, he coauthors Goldilocks RL with Ilia Mahrooghi and Emmanuel Abbe, an adaptive data-selection strategy that predicts reward variation across the model's rollouts for each candidate question. Across OpenMathReasoning and Polaris, the authors report that it reaches standard GRPO performance with up to 78% fewer optimization steps.

  8. As of October 2026

    Research Scientist at Apple

    As of October 2026, he is a Research Scientist at Apple Machine Learning Research, focused on reasoning capabilities in neural networks and large language models.

Career

  1. NowAppleResearch Scientist

Education

  1. 2025EPFLPhD
  2. Sharif University of Technology

From the biography.

  1. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Apple. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.
  2. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Apple. 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. Among employers with 30+ full-time cases, OpenAI leads primary technical role H-1B pay at $310,000, and Amazon, Google and Microsoft file 57% of all casesIncludes Apple. Across 440,669 certified US H-1B filings from 44 employers (FY2020–FY2026), OpenAI leads primary technical role base pay at $310,000 and xAI leads overall base pay at $290,000.
  4. The top 10 organizations employed 22% of people shaping AI in 2025: 51% of company researchers and 21% of university researchersIncludes Apple. We estimate the share of people shaping AI who worked at the top 10 organizations each year from 1955 to 2025, comparing concentration across companies, universities, and institutes and labs.

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 Aryo Lotfi yet. Send us a free-licensed photo.

  1. Learning to Reason with Neural Networks: Generalization, Unseen Data and Boolean Measures papers.neurips.cc. Checked 20 October 2026.
  2. ICML 2023 Awards icml.cc. Checked 20 October 2026.
  3. Apple AI/ML fellowship awarded to Aryo Lotfi - EPFL actu.epfl.ch. Checked 20 October 2026.
  4. Chain-of-Sketch: Enabling Global Visual Reasoning arxiv.org. Checked 20 October 2026.
  5. How Far Can Transformers Reason? The Globality Barrier and Inductive Scratchpad proceedings.neurips.cc. Checked 20 October 2026.
  6. RL for Reasoning by Adaptively Revealing Rationales arxiv.org. Checked 20 October 2026.
  7. Learning to Reason with Neural Networks: Principles and Methods - EPFL memento.epfl.ch. Checked 20 October 2026.
  8. Aryo Lotfi aryolotfi.com. Checked 20 October 2026.
  9. Goldilocks RL: Tuning Task Difficulty to Escape Sparse Rewards for Reasoning arxiv.org. Checked 20 October 2026.
  10. Publications - Apple Machine Learning Research machinelearning.apple.com. Checked 20 October 2026.