Projects
Public research write-ups, presented in full on-site. Each is a self-contained deep dive with its working, figures, and the parts that did not work.
-
How a network learns a finite group: irreps vs cosets, tested on a same-character-table pair
Does a grokked network compose group elements through their irreducible representations or through cosets? Two groups with identical character tables, built to tell the accounts apart, trace what divides them to representation type: the coset account adds nothing the irreps don't, while the real-versus-quaternionic split shows up twice over — in how fast each group learns and in a matrix-level readout.