Speaker
Jacky Hoi Tung Yip
Description
Fine, regular, and star triangulations (FRSTs) of 4-dimensional reflexive polytopes generate toric varieties, within which Calabi-Yau threefolds can be embedded as hypersurfaces. We use transformers---deep learning models originally developed for language modeling---to generate FRSTs of polytopes of varying sizes. Our method shows promising scalability with polytope size and can be continuously improved through self-improvement and priming strategies, laying the groundwork for a self-updating online database of Calabi-Yau manifolds.
