Preprint / scholarly article

Compute-Optimal AI via Image--EVI, Interior Bures--HK Control, and Fractal Dendritic Approximation (DIR)

K. Takahashi

Published
DOI
10.5281/zenodo.17576546

Abstract

Here alpha and beta are the empirical scaling exponents for parameters and tokens, is a robust image Lipschitz proxy, and (eta, rho) encode slack from acceptance tests.

Keywords

  • AI
  • large language models
  • geometry
  • information theory
  • compute reduction
  • scaling laws
  • entropy-transport
  • hellinger-kantorovich

Identifiers and source records