bibkey: “kumarbordelongershmanpehlevan2023grokking” authors: “Tanishq Kumar; Blake Bordelon; Samuel J. Gershman; Cengiz Pehlevan” year: 2023 title: “Grokking as the Transition from Lazy to Rich Training Dynamics” doi: null claim: “Grokking can arise from a transition from lazy (linearized) to rich (feature-learning) dynamics and can occur without regularization.” strata_touched: [] license: “citation-only” triage: “anchor” url: “https://arxiv.org/abs/2310.06110”

Grokking as the Transition from Lazy to Rich Training Dynamics

Verified locator

ICLR 2024; arXiv:2310.06110. Used only to delimit the fiber-law volume’s scope: its fixed-feature theorems do not cover the rich regime.

Declared identifiers: https://arxiv.org/abs/2310.06110.