Roman Soletskyi

Roman Soletskyi is a Ukrainian researcher based in Paris, France, currently working as a researcher at Mistral AI. He completed an MS in Physics at Ecole Normale Superieure – PSL (2022–2024) after undergraduate studies in physics at Moscow Institute of Physics and Technology (2020–2022), and won a gold medal at the International Physics Olympiad in 2017. His research spans AI safety and formal verification, machine learning theory, and applied deep learning. Notable work includes "Training Safe Neural Networks with Global SDP Bounds" (2024, co-authored with David Dalrymple of ARIA), which develops methods for training neural networks with formal safety guarantees using semidefinite programming, with applications to safe reinforcement learning policies; this work was supported by the Long-Term Future Fund and the Machine Learning Alignment Theory Scholars program. At Mistral AI he has contributed to projects including the Pixtral 12B multimodal model and research on variational inference theory.

No endorsements yet.

Personal Website

-

LinkedIn

LinkedIn

Twitter / X

-

LessWrong

-

EA Forum

-