This summer I attended two summer schools that map directly onto the two halves of my research: probabilistic machine learning, which underpins the movement primitives my robot skills are built on, and representation learning, which is where the foundation-model side of my work comes from.
ELLIS Probabilistic Machine Learning Summer School — Cambridge (July 14–18, 2025)
The Cambridge ELLIS Unit Summer School on Probabilistic Machine Learning at Pembroke College was a week-long deep dive: foundations first, then advanced probabilistic models, generative models, causality, and a closing day on probabilistic ML for accelerating science.
For me the causality day was the standout — causal reasoning is exactly what interactive robot learning needs when a refinement from a person should change why a motion is executed, not just how. Just as valuable were the conversations in between: a college full of researchers working on adjacent problems is the best code review one can get for half-formed ideas.
Certificate: ELLIS Cambridge 2025 (PDF, 0.7 MB)
OxML Summer School — Oxford: Representation Learning & Generative AI (August 7–10, 2025)
The Oxford Machine Learning Summer School (OxML) ran the MLx track on representation learning and generative AI, with speakers from Google DeepMind, Meta, Hugging Face and the Allen Institute for AI. The program covered language models and transformer architectures, visual representation learning, Bayesian deep learning, reinforcement learning, and geometric deep learning.
The sessions on how large models build their internal representations were the most useful for my work — when a robot uses a vision-language model to decide which skill fits a spoken request, the quality of those representations decides whether the robot understands “put it next to the blue box” correctly. The hands-on workshops made the trade-offs concrete in a way papers rarely do.
Certificate: OxML Oxford 2025 (PDF, 1.8 MB)