Lewis O’Donnell
I build time-series foundation models for fusion.
About
I’m a PhD student at UCL working on machine learning for nuclear fusion [FILL: group/centre name, supervisor if you want it listed]. My research focuses on time-series foundation models for plasma diagnostics: pretraining on large, heterogeneous tokamak signal data so that downstream tasks like [FILL: 1–2 concrete downstream tasks, e.g. disruption prediction / confinement-state classification] need far less labelled data.
I previously worked on geometric generative modelling during my MSc at [FILL: institution].
News
- Submitted “[FILL: workshop paper title]” to the ICML 2026 [FILL: workshop name] workshop.
- Released [FILL: benchmark name], an open-source benchmark for time-series foundation models.
- Started my PhD at UCL, working on machine learning for nuclear fusion.
Publications
-
[FILL: benchmark paper title]
-
[FILL: ICML workshop paper title]
Selected work
-
[FILL: benchmark name]
An open-source benchmark for evaluating time-series foundation models on [FILL: domain/tasks]. [FILL: one sentence on scale or what it enables].
-
Foundation models for plasma diagnostics (ongoing)
Pretraining transformer models on multi-machine tokamak diagnostic signals; the data is restricted, but all model and training code is released as I go.
-
Geometric generative modelling
[FILL: two sentences on the MSc project — what method, what problem.]