Hi, I’m Oscar Davis, a third year PhD student in Machine Learning at the University of Oxford, supervised by Prof Michael Bronstein, Dr İsmail Ceylan, and Dr Joey Bose. My funding is generously provided by both Project CETI and Intel. My work primarily focuses on generative modelling.

From November 2023 until February 2024, I interned with Microsoft Research, Cambridge, where I worked on diffusion models, stable diffusion, and briefly neural ODEs. We did some internal theoretical work on these, analysing their SDEs and ODEs. (Patented.)

I hold a BSc in Computer Science from EPFL with an exchange year at Imperial College London, and an MSc in Advanced Computer Science from the University of Oxford. I was also awarded the Tony Hoare Prize for the best MSc thesis of the year, Information Theoretic Perspectives on Graph Neural Networks.

Publications

Generalised Flow Maps for Few-Step Generative Modelling on Riemannian Manifolds

• Davis, O., Boffi, N., Albergo, M., Bronstein, M., Bose, J.
• NeurIPS 2025 FPI.
• arXiv: https://arxiv.org/abs/2510.21608.
• GitHub: https://github.com/olsdavis/gfm.

FORT: Forward-Only Regression Training of Normalizing Flows

• Rehman, D., Davis, O., Lu, J., Tang, J., Bronstein, M., Bengio, Y., Tong, A., Bose, J.
• ICML 2025 GenBio (best paper award).
• arXiv: https://arxiv.org/abs/2506.01158

SOAPI and SOAPIA

• SOAPIA: Vincoff, S.*, Davis, O.*, Ceylan, İ., Tong, A., Bose, J., Chatterjee, P.
• ICLR 2025 GEM (spotlight) and ICLR 2025 LMRL, respectively.

Fisher Flow Matching for Generative Modeling over Discrete Data

• Davis, O., Kessler, S., Petrache, M., Ceylan, İ., Bronstein, M., Bose, AJ.
• NeurIPS 2024.
• arXiv: https://arxiv.org/abs/2405.14664.
• GitHub: https://github.com/olsdavis/fisher-flow.

Projects

IMM re-implementation

As the Inductive Moment Matching paper came out in March, I originally re-implemented it for FORT from scratch. Although it has been a while, I thought it would be useful to share the code, as it is efficient, modular and organised. You can find it here: github.com/olsdavis/imm. Enjoy! :)