Yuhi Koike

Eective Regularisation from Loss-Landscape Geometry: A Unied Derivation of Direction-Dependent Grokking Dynamics

2026 · Zenodo (CERN European Organization for Nuclear Research)

Grokking is the phenomenon in which a neural network memorises its training data and then, after a prolonged delay, suddenly generalises. Two problems remain open: grokking can occur even at weight-decay strength β=0, and a …