Meet the Fellow: Simone Bombari
This entry is part of our Meet the Fellow blog series, which introduces and highlights incoming Faculty Fellows at CDS.
Meet incoming CDS Faculty Fellow Simone Bombari, who is joining CDS this fall after completing his PhD at the Institute of Science and Technology Austria (ISTA), advised by ISTA Professor Marco Mondelli.
Bombari came to machine learning by way of physics. He studied at Scuola Normale Superiore and the University of Pisa, working on statistical physics and quantum information, before the modeling and statistics underneath that work started to interest him more than the physics itself. “It was a slow approach away from physics,” Bombari said. By the time he reached his master’s thesis, he was working on deep learning, and his PhD settled into learning theory and the statistics of how models behave.
Get NYU Center for Data Science’s stories in your inbox
Join Medium for free to get updates from this writer.
His recent research centers on a question that sounds almost philosophical until the math takes over: what separates learning from memorizing? Bombari studies deep learning theory and high-dimensional probability, with much of his attention on trustworthy machine learning, particularly privacy. This last thread started during a six-month internship at Amazon early in his PhD, where he was assigned to a product privacy team. “It has the flavor of information theory and statistics that I like,” Bombari said. He picked up the topic there, then returned to ISTA wanting to chase the mathematical questions underneath it.
This field looks at cases where a model not only learns from data, but also learns some of them by heart, whether a user’s private information or a copyrighted piece of work. Bombari pointed to the lawsuits against AI labs over LLMs trained on newspaper and books archives. His paper “Privacy for Free in the Overparameterized Regime,” published in the Proceedings of the National Academy of Sciences with Mondelli, works out cases where large models can be successfully trained while preserving privacy. “What does it mean to learn? What does it mean to memorize?” Bombari asked. “Math and statistics can capture this, up to a point.”
Along the way he has held a Google PhD Fellowship since the fall of 2024, spent time as an NLP quant researcher at G-Research, and visited the University of Southern California in 2025 to work with the groups of Yizhe Zhu and Mahdi Soltanolkotabi.
At NYU, Bombari expects memorization and privacy to keep occupying him, though he described his plans with deliberate openness. He wants roughly half his work anchored in what he already knows he can do, and the other half free to follow wherever the research community pulls him. He mentioned the density of machine learning researchers across NYU as a reason he is confident he will not run out of problems. “Maybe I’ll end up working on AI for the natural sciences,” Bombari said. “You never know.”
To view all of our current faculty fellows, please visit the CDS Faculty Fellows page on our website.
By Stephen Thomas
