How they use Nebius AI Cloud

The London Institute for Mathematical Sciences uses Nebius compute for research in mathematics and theoretical physics. A 2024 researcher article describes donated computing time under a research grant, including H100s and other GPUs. Nebius’s July 2024 digest says it hosted multiple LIMS projects. The researcher article also describes plans to use additional compute to analyze UKRI grant abstracts and arXiv papers, cluster them by subject and map research fields.

A Nebius customer story describes LIMS training and testing BERT-based models on rule-abstraction problems. Researchers used elementary cellular automata to investigate whether transformer models can abstract rules and how neural networks learn them. The study reported validation accuracy above 0.93, including testing on rules absent from the training set. In its r3s24T20 experiment, the model trained on about a million different rules. The story lists Compute Cloud and Virtual Private Cloud as services used.

The 2024 article described plans to use Nebius compute for Evgeny Sobko’s research on quantum integrable models. A June 2026 Nebius Science feature describes Sobko and his collaborators using neural networks, supported by Nebius cloud infrastructure, to discover thousands of new systems defined by the Yang-Baxter equation within the class they studied. LIMS also thanked Nebius in 2026 for providing compute for his AI-driven discovery work. An August 2026 Nebius article names the institute among Research Grants participants. Nebius names the institute as a customer. LIMS also identifies Nebius as sponsor of its monthly AI for Mathematical Sciences seminar series. LIMS announced that it was hosting a free, in-person, 14-week course developed by Nebius Academy on using AI in mathematical research.