Machine Learning Foundation Models for Astronomy (Likely to change, still in early stages of project and goals will evolve.)

(Likely to change, still in early stages of project and goals will evolve.)

Develop transformer-based machine learning foundation models for astronomy as part of the Polymathic AI collaboration. Initially this will be applied to large galaxy surveys (currently working on DESI data) with a vision to develop methods that address the challenges of future surveys such as LSST. Possible applications include photometric redshift prediction, and the model may generalise effectively to high-z prediction with fine tuning despite the low data regime.

Currently working on how to create joint representations of multimodal inputs. Other issues that may be tackled include data heterogeneity (e.g. making models robust to different PSF, noise, etc. in data from different instruments) and ML interpretability.