매일 아침, 어제의 AI를 한 통으로 정리해 보내드립니다메일로 받아보기

METAL LAB

Skaling: Chinchilla's Exponents Meet Kaplan's Coupling

arXiv:2608.072222026-08-10

arXiv:2608.07222v1 Announce Type: new Abstract: Neural scaling laws are foundational for language model development, yet standard formulations systematically under- and overestimate loss at data-scarce and overtraining extremes. This failure originates in the underlying assumption that model size and training data impact the loss independently. To address this, we introduce the Skaling law, a generalized functional form that couples model capacity and data through a single interaction exponent.

저자 · Mathurin Videau, Badr Youbi-Idrissi, David Lopez-Paz, Kartik Ahuja

arXiv에서 원문 보기