Hardy-Szegő Point Processes: Large Deviations and Strong Szegő Asymptotics

What the math gives to ML

The paper provides an explicit conformally invariant determinantal point process whose kernel produces strong repulsion while remaining translation-invariant in the horizontal coordinate. The transferable asset is not the Hardy–Szegő zero process itself, but the positive-definite complex kernel and its determinant-based diversity geometry, which can turn token selection or attention routing into a repulsive subset-selection problem. A practical first transfer is a quality-weighted determinantal router that favors useful tokens while suppressing redundant tokens, with greedy Schur-complement selection or Nyström approximations controlling the determinant cost.