CaMLSysUniversity of Cambridge

Cambridge
Machine Learning
Systems Lab.

The Cambridge Machine Learning Systems Lab dodo emblem

Research

Wide-Area Distributed Training

We develop algorithms and systems to train large-scale models across geographically distributed, heterogeneous infrastructure.

Agentic AI and Self-Improvement

We build AI agents that solve complex tasks, learn from experience and improve their through experimentation and feedback.

AI Safety

As systems become more capable and autonomous, we study how to make AI reliable, controllable and accountable.

Efficiency and Sustainability

We improve computing efficiency at the hardware level and explore how datacenters can reduce their environmental impact.

News

  • Our paper The Red Queen Gödel Machine: Co-Evolving Agents and Their Evaluators explores why self-improving systems need evaluations that improve alongside them.

  • Our paper LoRDO was presented at ICML in Seoul. It combines low-rank optimization with infrequent communication to support wide-area distributed training.

  • Cambridge joins SOFAIR, led by David Barber at UCL in partnership with Oxford and Edinburgh. Our contribution focuses on training frontier AI models across wide-area networks of heterogeneous compute.

  • Two papers at ICLR in Rio de Janeiro: MT-DAO and DES-LOC, advancing wide-area distributed training of large models.

All newsLess news