·reading

Grokking: Generalization Beyond Overfitting

The paper where validation accuracy jumps from chance to perfect long after the training loss flatlined — generalization as a delayed phase change.

paper · finished
Alethea Power, Yuri Burda, Harri Edwards, Igor Babuschkin, Vedant Misra
arXiv:2201.02177
source ↗

Full title: Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets. The figure does the arguing — training accuracy saturates near 10³ steps, validation sits at chance for two more orders of magnitude, and then snaps to 100%.

What makes it a favorite is not the phenomenon but what the phenomenon costs you. Early stopping on validation loss, the most standard practice there is, would have killed every one of these runs long before the interesting thing happened. The training curve was not a proxy for the thing being learned; the network was still reorganizing under a flat loss.

Small algorithmic datasets are a cheap testbed for exactly this — a single GPU, a clean ground truth, and a generalization event you can watch arrive.

Neighborhood

Related

Language Models are Few-Shot LearnersLanguage Models are Few...Predictive General IntelligencePredictive General Inte...Reward is EnoughReward is EnoughThe Bitter LessonThe Bitter LessonThe Unreasonable Effectiveness of *The Unreasonable Effect...Reaching for the IntangibleReaching for the Intang...Generalization -- Fast and Slow (incomplete)Generalization -- Fast and ...On the Measure of IntelligenceOn the Measure of Intellige...What is intelligence?What is intelligence?Grokking: Generalization Beyond...