# Nadav Cohen

> Professor of Computer Science & AI at Tel Aviv University, and Co-Founder & CTO at Imubit. Researcher in the foundations of deep learning, world models and physical AI.

Nadav Cohen leads research spanning the mathematical foundations of deep learning (neural networks) through to deployment in the physical world. His academic group develops theory and conducts systematic experiments to yield principled algorithms. His industrial labs build world models for decision making (e.g. via reinforcement learning) in physical AI, with systems in production at global scale.

He earned dual BSc degrees in Electrical Engineering and Mathematics (both summa cum laude) from the Technion Excellence Program, a PhD from the Hebrew University of Jerusalem (supervisor: Prof. Amnon Shashua), and was a Postdoctoral Fellow at the Institute for Advanced Study in Princeton (hosts: Prof. Sanjeev Arora and Prof. Elad Hazan).

## Website Pages

- [Homepage](https://cohennadav.com/index.html): Overview, affiliations, and news
- [Biography](https://cohennadav.com/bio.html): Full biography and CV
- [Publications](https://cohennadav.com/publications.html): Full list of papers
- [Research Group](https://cohennadav.com/group.html): Current and former students
- [Selected Talks](https://cohennadav.com/talks.html): Lecture slides and talk recordings
- [Blog Posts](https://cohennadav.com/blogs.html): Posts on Off the Convex Path blog
- [Teaching](https://cohennadav.com/teaching.html): Courses taught
- [Patents](https://cohennadav.com/patents.html): Patents
- [Media](https://cohennadav.com/media.html): Media appearances and interviews
- [CV](https://cohennadav.github.io/files/cv/cv.pdf): Full curriculum vitae (PDF)

## Research Topics

- Theory of deep learning (neural networks): expressiveness, optimization, generalization, implicit regularization
- Tensor decompositions and their connections to deep learning architectures
- Implicit bias of gradient descent in matrix/tensor factorization and neural networks
- Reinforcement learning theory, world models, physical AI
- Structured state space models (SSMs / Mamba)
- Connections between deep learning and quantum entanglement / tensor networks
- Agentic AI safety

## Selected Publications

- [Why Does Agentic Safety Fail to Generalize Across Tasks?](https://arxiv.org/abs/2605.06992) — Slutzky, Alexander, Slor, Nagel, Cohen. arXiv 2026.
- [Outcome-Based RL Provably Leads Transformers to Reason, but Only With the Right Data](https://arxiv.org/abs/2601.15158) — Ran-Milo, Alexander, Mendel, Cohen. arXiv 2026.
- [Do Neural Networks Need Gradient Descent to Generalize? A Theoretical Study](https://openreview.net/forum?id=lbjKWBzK9k) — Alexander, Slutzky, Ran-Milo, Cohen. NeurIPS 2025.
- [The Implicit Bias of Structured State Space Models Can Be Poisoned with Clean Labels](https://openreview.net/forum?id=3UaItHVjyE) — Slutzky, Alexander, Razin, Cohen. NeurIPS 2025 Spotlight (top 3%).
- [What Makes Data Suitable for a Locally Connected Neural Network? A Necessary and Sufficient Condition Based on Quantum Entanglement](https://openreview.net/forum?id=4aIpgq1nuI) — Alexander, De La Vega, Razin, Cohen. NeurIPS 2023 Spotlight (top 3%).
- [Implicit Regularization in Deep Learning May Not Be Explainable by Norms](https://arxiv.org/abs/2005.06398) — Razin, Cohen. NeurIPS 2020.
- [Implicit Regularization in Deep Matrix Factorization](https://papers.nips.cc/paper/8960-implicit-regularization-in-deep-matrix-factorization) — Arora, Cohen, Hu, Luo. NeurIPS 2019 Spotlight (top 3%).
- [On the Optimization of Deep Networks: Implicit Acceleration by Overparameterization](http://proceedings.mlr.press/v80/arora18a.html) — Arora, Cohen, Hazan. ICML 2018.
- [On the Expressive Power of Deep Learning: A Tensor Analysis](https://proceedings.mlr.press/v49/cohen16.html) — Cohen, Sharir, Shashua. COLT 2016.
- [Lecture Notes on Linear Neural Networks](https://arxiv.org/abs/2408.13767) — Cohen, Razin. 2024.

## Honors and Awards

- Senior Member, US National Academy of Inventors (2026)
- Israel AI Safety Research Prize (2026)
- ERC Starting Grant (2024)
- Google Research Scholar Award (2021)
- ELLIS Scholar (2020)
- Rothschild Postdoctoral Fellowship (2017)
- Zuckerman Postdoctoral Fellowship (2017)
- Google Doctoral Fellowship in Machine Learning (2015)

## Social and External Profiles

- [LinkedIn](https://www.linkedin.com/in/cohennadav/)
- [X / Twitter](https://x.com/nadavcohen)
- [YouTube](https://www.youtube.com/@cohennadav-ai)
- [Google Scholar](https://scholar.google.com/citations?user=AfLwLQ0AAAAJ&sortby=pubdate)