CV
A slightly longer version of the story on the front page.
General Information
| Full Name | Itay Lavie |
| Position | PhD Student, Applied Mathematics, Harvard University |
| Research Areas | Theory of deep learning, statistical physics of learning |
| Languages | Hebrew, English |
Education
-
2025 - present PhD, Applied Mathematics
Harvard University - Advised by Prof. Cengiz Pehlevan.
- GPA: 4.0/4.0.
- Dynamical mean-field theories of implicit bias and scaling laws for modern optimizers.
-
2022 - 2024 M.Sc., Physics
Hebrew University of Jerusalem - Advised by Prof. Zohar Ringel.
- Thesis: "Towards Understanding Inductive Bias in Transformers: A View from Infinity".
- GPA: 97/100 — Magna cum laude.
-
2019 - 2022 B.Sc., Physics (major), Philosophy (minor)
Hebrew University of Jerusalem - "Amirim" honors program.
- GPA: 97/100 — Summa cum laude.
Research Experience
-
2025 - present Graduate Student Researcher
Pehlevan Group, Harvard University - Developing dynamical mean-field theories of implicit bias and scaling laws for modern optimizers.
-
2024 Research Intern, Decision Optimization Group
IBM Research - Deep learning models for combinatorial optimization.
- A theoretical framework for out-of-distribution generalization in combinatorial optimization.
-
2023 - 2024 Helmholtz Visiting Researcher
Theory of Multi-Scale Neuronal Networks group (Prof. Moritz Helias), Forschungszentrum Jülich - Representation learning and feature learning in transformers; collaboration ongoing.
-
2022 - 2024 Graduate Student Researcher
Ringel Group, Hebrew University of Jerusalem - Infinite-width theory of transformer inductive bias.
- Sample-complexity bounds for real-world datasets via their symmetric spectral structure.
-
2020 - 2022 Undergraduate Student Researcher
Hochberg & Kuflik Group, Racah Institute of Physics, Hebrew University of Jerusalem - Bachelor's thesis: "Roadmap to Thermal Dark Matter Beyond the WIMP Unitarity Bound", published in Physical Review Letters.
- Out-of-equilibrium dynamics in the early universe as a route to the hierarchy problem.
Talks & Conferences
-
2026 Poster presentation & participant, Theoretical Physics for Artificial Intelligence
Aspen Center for Physics -
2025 Invited talk, Machine Learning Seminar, Faculty of Electrical Engineering
Technion — Israel Institute of Technology -
2024 Participant & poster presentation, Theory of Machine Learning School
Princeton University -
2024 Poster presentation, 41st International Conference on Machine Learning (ICML)
ICML 2024 -
2024 Contributed talk, Israeli Physics for Deep Learning Conference
Israel -
2024 Poster presentation, Workshop on Mathematical and Empirical Understanding of Foundation Models
ICLR 2024
Community Service
-
2026 Reviewer, 43rd International Conference on Machine Learning
ICML 2026 -
2024 Reviewer, 13th International Conference on Learning Representations
ICLR 2025 -
2024 Organizer, Phys4ML social
ICML 2024 - Brought together 200+ attendees working at the intersection of physics and machine learning.
Honors and Awards
-
2026 - Google DeepMind Aspen Center for Physics Travel Grant
-
2025 - Open Philanthropy Career Development and Transition Grant — supporting dedicated time on AI safety and interpretability
-
2023 - Helmholtz Visiting Researcher Grant, Helmholtz Information and Data Science Academy
- M.Sc. Dean's List, Faculty of Sciences, Hebrew University of Jerusalem
-
2021 - 2022 - M.Sc. Direct Track Fellowship, Racah Institute of Physics, Hebrew University of Jerusalem
-
2020 - 2022 - "Amirim" Scholarship — full tuition, awarded to the top 3% of the Faculty of Sciences student body
-
2021 - Amos de-Shalit summer program for selected students in physics, Weizmann Institute of Science
- Research workshop for outstanding undergraduate physics students, Racah Institute of Physics
-
2020 - 2022 - B.Sc. Dean's List, Faculty of Sciences, Hebrew University of Jerusalem (2020, 2021, 2022)
Teaching & Mentoring
-
2024 Lab instructor, Physics Lab III: Advanced Experiments in Quantum and Modern Physics
Hebrew University of Jerusalem -
2023 Lab instructor, Physics Lab I & II: Waves, Dynamical Systems and Materials Properties
Hebrew University of Jerusalem (Talpiot) -
2024 Tutor, Electricity and Magnetism
Hebrew University of Jerusalem - Mentored a student from a disadvantaged background through the university's IDEA unit.
-
2020 - 2021 Tutor, Mathematical Methods I & II
Hebrew University of Jerusalem - Mentored a student from a disadvantaged background through the university's IDEA unit.
Academic Interests
-
Theory of deep learning
- Inductive and spectral bias — which functions a model finds easy to learn, and why.
- Generalization and scaling laws; implicit bias of modern optimizers.
-
Transformers and attention
- In-context learning and the emergence of induction heads.
- Feature learning and phase transitions during training.
-
Statistical physics of learning
- Gaussian processes, kernel methods, and infinite-width limits.
- Mean-field theory and the representation theory of symmetry groups.
-
Earlier work — high-energy phenomenology
- Thermal dark matter and freeze-out beyond the WIMP unitarity bound.