Lorenzo Sani

I am a Research Scientist at Flower Labs, where I work on federated and distributed training of foundation models. In parallel I am finishing my PhD in Computer Science at the University of Cambridge, supervised by Prof. Nicholas D. Lane in the CaMLSys group at the Computer Laboratory.

My research asks a simple question: how do we train large models when the data and the compute are spread across the world and cannot be pooled? That has led me through federated pre-training of language models — the Photon system and the DEPT method — and, more recently, to communication-efficient distributed optimizers such as DES-LOC, MT-DAO, and LoRDO, and to sparse and modular architectures like FoMoE.

Before Cambridge, I read Physics at the University of Bologna: a Bachelor’s in 2019 with a thesis on the fine structure of hydrogen-like atoms, and a Master’s in Applied Physics in 2021 with a thesis on unsupervised clustering of MDS data using federated learning, supervised by Prof. Enrico Giampieri and Prof. Gastone Castellani. During that time I contributed to the GenoMed4All project — including Deliverable 6.2 on federated learning for clinical data — and began collaborating with CaMLSys on the Flower framework, now the most widely used open-source federated learning framework.