The paper investigates how overparameterized neural networks can fit training data in many different ways yet generalize differently. By applying short, fixed‑duration weight‑decay pulses during the grokking plateau, the authors show that early in training the pulses produce unordered shifts in generalization time, but later a clear dose ordering emerges: larger weight‑decay increases lead to earlier generalization, while larger decreases delay it. This ordering appears before visible generalization and persists even as test‑loss barriers collapse, a phenomenon the authors term the canalization of function selection.
By Yiming Lin
arXiv:2607. 23967v1 Announce Type: new Abstract: Delayed generalization, or grokking, remains poorly understood despite extensive empirical study.
By Taeyoung Kim
arXiv:2601. 19791v4 Announce Type: replace Abstract: We study grokking, the onset of generalization long after overfitting, in a classical ridge regression setting.
By Mingyue Xu, Gal Vardi, Itay Safran
arXiv:2609.07755v1 Announce Type: new
Abstract: Understanding generalization remains a central challenge in machine learning because it requires jointly considering data, architecture, and training d...
By Yuqing Wang, Ioannis G. Kevrekidis, Mikhail Belkin
arXiv:2607. 20552v1 Announce Type: new Abstract: Grokking -- the delayed generalization of neural networks long after they have memorized their training data -- wastes thousands of training epochs and is notoriously unpredictable.
By Chitraansh Pandey
arXiv:2602.01718v2 Announce Type: replace
Abstract: Predicting generalization from quantities available before target-test evaluation remains a central challenge in deep learning. The systematic benc...
By Sora Nakai, Youssef Fadhloun, Kacem Mathlouthi, Kotaro Yoshida, Ganesh Talluri, Ioannis Mitliagkas, Hiroki Naganuma
The study investigates the delayed transition from memorization to generalization—known as grokking—in two‑hidden‑layer MLPs trained on modular arithmetic. By exploring 384 hyperparameter configurations, the authors derive a power‑law scaling relation for the onset time of generalization, showing that data complexity dominates over model capacity. A clear phase boundary at weight decay around 1.0 separates grokking from non‑grokking regimes, and weight norm trajectories indicate implicit regularization during the transition.
By Anish Kataria
The paper presents a spectral theory explaining the phenomenon of grokking, where an initial fit to training data is followed by a delayed improvement in generalization. It shows that for homogeneous networks trained with squared loss and L₂ weight decay, residuals after memorization influence the neural tangent kernel (NTK) dynamics, leading to a transition from lazy to rich learning. The theory predicts that grokking timescales depend on the product of learning rate and weight decay, and that stronger decay can halt fitting, with empirical validation on modular addition tasks using MLPs and Transformers.
By Lenz Pracher, Pascal de Jong, Oskar Lieshaus, Alan Jeffares, Steffen Rulands
arXiv:2606. 30388v1 Announce Type: cross 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.
By R\'ois\'in Luo, Christian Gagn\'e, Jonas Ngnaw\'e, Ihsan Ullah, Karyn Morrissey
arXiv:2507. 01598v5 Announce Type: replace Abstract: Muon, a recently proposed optimizer that leverages the inherent matrix structure of neural network parameters, has demonstrated strong empirical performance, indicating its potential as a successor to standard optimizers such as AdamW.
By Naoki Sato, Hiroki Naganuma, Hideaki Iiduka
ExpTest is an autonomous learning‑rate controller that uses the training loss curve as an online signal to perform sequential statistical tests on theoretically motivated windows, detecting convergent behavior and triggering learning‑rate reductions. It combines a covariance‑based initial learning‑rate estimate, curvature‑motivated window sizing, and a two‑phase test‑driven decay, relying on the approximately exponential decay predicted under linearized network dynamics. Experiments on regression, classification, forecasting, and natural‑language tasks across various architectures show that ExpTest achieves competitive performance compared to hand‑tuned SGD baselines and recent learning‑rate‑free methods, without requiring manual initial learning‑rate selection or predefined scheduling.
By Zan Chaudhry, Naoko Mizuno
The paper studies how different optimizers perform as training duration (overtraining) increases, focusing on matrix‑preconditioned methods (Muon, SOAP) and a momentum‑scheduled method (ADANA) compared to AdamW. Across models ranging from 51M to 253M parameters and overtraining factors up to 256×, the authors find that optimal learning‑rate schedules, weight‑decay coefficients, and memory settings shift with horizon, and that ADANA consistently outperforms AdamW, especially with log‑time weight decay and momentum cooldown. Muon and SOAP maintain roughly constant token‑efficiency advantages, with SOAP potentially improving at the highest overtraining levels.
By Katie Everett, Shikai Qiu