Grokking as a Falsifiable Finite-Size Transition
This work addresses the need for rigorous testing of phase-transition claims in machine learning, offering a method to move beyond analogies, though it is incremental in refining existing concepts.
The paper tackled the problem of verifying grokking as a genuine phase transition by providing falsifiable finite-size inputs, using group order as an extensive variable and a spectral contrast as an order parameter, and found strong evidence against a smooth-crossover interpretation with a ΔAIC of 16.8.
Grokking -- the delayed onset of generalization after early memorization -- is often described with phase-transition language, but that claim has lacked falsifiable finite-size inputs. Here we supply those inputs by treating the group order $p$ of $\mathbb{Z}_p$ as an admissible extensive variable and a held-out spectral head-tail contrast as a representation-level order parameter, then apply a condensed-matter-style diagnostic chain to coarse-grid sweeps and a dense near-critical addition audit. Binder-like crossings reveal a shared finite-size boundary, and susceptibility comparison strongly disfavors a smooth-crossover interpretation ($Î\mathrm{AIC}=16.8$ in the near-critical audit). Phase-transition language in grokking can therefore be tested as a quantitative finite-size claim rather than invoked as analogy alone, although the transition order remains unresolved at present.