arXiv stat.ML
· Papers
A Stochastic–Geometric Theory of Scaling Laws in Grokking
arXiv:2606.30388v3 Announce Type: replace Abstract: Delayed generalization (ie~grokking) refers to the phenomenon in which a neural network fits its training data early in training but only begins to generalize after a prolonged delay, often through an abrupt transition. Despite extensive empirical study, its underlyi