A paper led by Zak Shumaylov and Peter Zaika (both at the University of Cambridge, UK) will be presented at ICLR 2025.
Z Shumaylov, P Zaika, J Rowbottom, F Sherry, M Weber, and CB Schönlieb. Lie algebra canonicalization: Equivariant neural operators under
arbitrary Lie groups ICLR 2025
This paper introduces LieLAC, a canonicalization approach for enforcing equivariance in neural operator based PDE solver. LieLAC is applicable even without full group knowledge and enables robust symmetry alignment of pre-trained models.
A second paper from the AstroAI Initiative at the Center for Astrophysics | Harvard & Smithonian will be presented at the ICLR Re-Align Workshop.
JR Martínez-Galarza, NO Pinciroli Vago, S Raval, C Cuesta-Lazaro, M Weber, D Alvarez-Melis, A Accomazzi, C Garraffo, J Knutson, R Thill, CB Green, I Ahangama. Augmenting X-ray Astronomical Representations with Scientific Knowledge through Contrastive Learning. ICLR Workshop on Representational Alignment (Re-Align)
This paper uses physics-informed alignment of X-ray data and LLM-based text embeddings via contrastive learning for regression tasks and anomaly detection in astrophysics.