Alternative Basis

"Alternative basis" research explores methods for representing data and solving problems using unconventional approaches, moving beyond standard techniques. Current work focuses on developing more expressive graph neural networks through optimized homomorphism counts, improving Gaussian process efficiency with kernel packets, and employing novel frameworks for physics-informed learning and text preference prediction. These advancements aim to enhance model expressiveness, computational efficiency, and interpretability across diverse fields, impacting areas such as fluid dynamics, natural language processing, and machine learning for scientific discovery.