I'm a postdoctoral researcher working at the intersection of astrophysics and machine learning, focusing on how complex structures emerge and evolve in the Universe. My research centers on hierarchical clustering, metric learning, and diffeomorphic transformations for modeling structure in high-dimensional astrophysical data.
I'm broadly interested in:
- Machine learning methods for structure formation, representation learning, and uncertainty modeling
- Developing open-source tools that make scientific ML more interpretable and reproducible
- Cross-disciplinary applications of clustering and flow-based modeling
(Clusters in simulated Milky Way-like galaxy found with the FuzzyCat + AstroLink pipeline. Watch more movies like this here.)
| Project | Description |
|---|---|
| AstroLink | A hierarchical clustering algorithm built for large astrophysical datasets. Provides robust and interpretable structure with minimal parameter tuning. |
| FuzzyCat | An unsupervised and data-blind algorithm built to account for the effects of change-processes on clusters. Useful for propagating uncertainties, constructing time-invariant labels, and abstracting over model choices. |
| Fishereyes | A JAX-based framework for uncertainty-aware representation learning of heteroskedastic data. Provides a multi-dimensional precision-weighted embedding of the data into Euclidean space where measuring distances is easy. |
Languages: Python, Numba, NumPy, SciPy
Machine Learning: JAX, PyTorch, scikit-learn, Flow-based models, Neural ODEs
Astronomical Tooling: Astropy, HEALPix, Gaia DR3 data handling
Development: Git, pytest, Jupyter, documentation automation (Sphinx), CI/CD workflows
I’m always open to discussions or collaborations on topics related to machine learning for astrophysics, structure formation, or scientific representation learning.
If you’d like to get in touch, feel free to open an issue on any of my repositories or reach out directly by email.
📧 william.oliver@iwr.uni-heidelberg.de | william.hardie.oliver@gmail.com