ΠΠ°Π»Π΅Π΅Π‘Π»Π΅Π΄ΡΡΡΠ΅Π΅ ΡΡΠΊΠΎΠ²ΠΎΠ΄ΡΡΠ²ΠΎ
Π Π΅Π³ΡΠ»ΡΡΠΈΠ·Π°ΡΠΈΡ
ΠΡΠ½ΠΎΠ²Ρ
Π Π£ΠΠΠΠΠΠ‘Π’ΠΠ ΠΠ ΠΠ‘ΠΠΠΠΠ
Grokking describes studied cases where held-out generalization improves well after a network has fit its training examples.
ΠΡΠΎΡΠ»ΡΡΠ°ΠΉΡΠ΅ ΡΡΠΎ ΡΡΠΊΠΎΠ²ΠΎΠ΄ΡΡΠ²ΠΎΠΠ·Π²ΡΡΠ΅Π½ΠΎ ΠΠ Β· ~3 min
Small algorithmic tasks provide clear experiments, but neither a validation plateau nor longer training guarantees that this transition will occur.
The original grokking paper, publicly submitted in January 2022, studies small algorithmically generated datasets. In some experiments, a network fits the training data while held-out performance remains poor, then generalizes after substantially more optimization. This separates fitting observed examples from learning behavior that transfers to unseen examples of the underlying task. It does not mean that every model has near-zero validation accuracy during an initial plateau or that memorization must always disappear completely. The original study varies dataset sizes and optimization choices, including weight decay. Its findings concern those setups rather than a unique universal explanation of generalization. Progress Measures for Grokking via Mechanistic Interpretability investigates small transformers trained on modular addition. The studied algorithm uses Fourier structure and trigonometric identities to represent addition through circular relationships. The researchers examine activations and weights and use Fourier-space ablations to test their account. They describe memorization, circuit formation, and cleanup phases, with generalizing structure emerging before the visible jump in test accuracy. Weight decay removes memorization components during the studied cleanup process. This is more specific than a story in which the entire algorithm appears suddenly at the accuracy jump. Modular tasks offer exact labels: for example, adding six and five modulo seven gives four because eleven leaves remainder four after division by seven. The original paper also reports increasing optimization requirements as training datasets become smaller. Such controlled experiments motivate further research, but do not establish that a large deployed model will eventually improve if trained indefinitely.
Plotting training and held-out accuracy on a modular-arithmetic experiment.
Comparing regularization settings on a defined small algorithmic dataset.
Inspecting the Fourier-based circuit studied in a small transformer.
Varying training-set size while recording the optimization needed for held-out performance.
Controlled experiments can test how dataset size, architecture, regularization, and optimization affect delayed generalization. Progress measures based on a verified circuit may reveal behavior hidden by an overall accuracy curve, though scaling such analysis to larger models remains an open question. Researchers should distinguish a measured mechanism in a small task from a proposed explanation for every model. Practical stopping decisions still need evaluation, resource constraints, and evidence of progress; the existence of grokking provides a reason to investigate some plateaus rather than a rule to continue all training indefinitely.
ΠΠ΅ΡΠΏΠ»Π°ΡΠ½Π°Ρ ΡΠ°ΡΡΡΠ»ΠΊΠ°
Π’ΡΠΈ ΠΏΡΠΎΠ²Π΅ΡΠ΅Π½Π½ΡΡ AI-ΠΈΡΡΠΎΡΠΈΠΈ ΠΊΠ°ΠΆΠ΄ΠΎΠ΅ ΡΡΡΠΎ Π±ΡΠ΄Π½Π΅Π³ΠΎ Π΄Π½Ρ, Π½Π°ΠΏΠΈΡΠ°Π½Π½ΡΠ΅ ΠΏΡΠΎΡΡΡΠΌ Π°Π½Π³Π»ΠΈΠΉΡΠΊΠΈΠΌ. ΠΠ΅ΡΠΏΠ»Π°ΡΠ½ΠΎ Π½Π°Π²ΡΠ΅Π³Π΄Π°, Π±Π΅Π· ΡΠ΅ΠΊΠ»Π°ΠΌΡ.
ΠΠ΄Π½ΠΎ ΡΠ»Π΅ΠΊΡΡΠΎΠ½Π½ΠΎΠ΅ ΠΏΠΈΡΡΠΌΠΎ ΠΊΠ°ΠΆΠ΄ΡΠΉ Π±ΡΠ΄Π½ΠΈΠΉ Π΄Π΅Π½Ρ. ΠΡΠΏΠΈΡΠ°ΡΡΡΡ Π² ΠΎΠ΄ΠΈΠ½ ΠΊΠ»ΠΈΠΊ. ΠΡ Π½ΠΈΠΊΠΎΠ³Π΄Π° Π½Π΅ ΠΏΡΠΎΠ΄Π°Π΅ΠΌ ΠΈ Π½Π΅ ΠΏΠ΅ΡΠ΅Π΄Π°Π΅ΠΌ Π²Π°Ρ Π°Π΄ΡΠ΅Ρ.
ΠΡΠΎΠ²Π΅ΡΡΡΠ΅ ΡΠ΅Π±Ρ
ΠΠ³Π½ΠΎΠ²Π΅Π½Π½Π°Ρ ΠΎΠ±ΡΠ°ΡΠ½Π°Ρ ΡΠ²ΡΠ·Ρ ΠΏΠΎ ΠΊΠ°ΠΆΠ΄ΠΎΠΌΡ ΠΎΡΠ²Π΅ΡΡ ΠΈ ΠΎΠ±ΡΠΈΠΉ ΡΠ΅ΡΡΠΈΡΠΈΠΊΠ°Ρ Ρ ΠΏΡΠΎΠ²Π΅ΡΡΠ΅ΠΌΡΠΌ ΠΈΠ΄Π΅Π½ΡΠΈΡΠΈΠΊΠ°ΡΠΎΡΠΎΠΌ ΠΏΠΎΡΠ»Π΅ ΠΏΡΠΎΡ ΠΎΠΆΠ΄Π΅Π½ΠΈΡ ΠΊΡΡΡΠ°.
ΠΠΎΠ΄Π΄Π΅ΡΠΆΠΈΡΠ΅ Π±Π΅ΡΠΏΠ»Π°ΡΠ½ΠΎΠ΅ ΠΎΠ±ΡΠ°Π·ΠΎΠ²Π°Π½ΠΈΠ΅ Π² ΠΎΠ±Π»Π°ΡΡΠΈ ΠΈΡΠΊΡΡΡΡΠ²Π΅Π½Π½ΠΎΠ³ΠΎ ΠΈΠ½ΡΠ΅Π»Π»Π΅ΠΊΡΠ°. AI Understanding β ΡΡΠΎ Π½Π΅ΠΊΠΎΠΌΠΌΠ΅ΡΡΠ΅ΡΠΊΠ°Ρ ΠΎΡΠ³Π°Π½ΠΈΠ·Π°ΡΠΈΡ 501(c)(3). ΠΠΈΠΊΠ°ΠΊΠΎΠΉ ΡΠ΅ΠΊΠ»Π°ΠΌΡ ΠΈ ΠΏΠ»Π°ΡΠ½ΠΎΠ³ΠΎ Π΄ΠΎΡΡΡΠΏΠ°. Π‘Π΄Π΅Π»Π°ΡΡ ΠΏΠΎΠΆΠ΅ΡΡΠ²ΠΎΠ²Π°Π½ΠΈΠ΅
Grokking describes studied cases where held-out generalization improves well after a network has fit its training examples. Small algorithmic tasks provide clear experiments, but neither a validation plateau nor longer training guarantees that this transition will occur.
The delay concerns generalization beyond observed examples under the studied setup.
The public paper was submitted in January 2022 and studies small algorithmic datasets.
The original study investigates weight decay and other optimization choices; no single universal cause is established.
Analysis and Fourier ablations support the particular studied algorithm.
The generalizing circuit forms before the later cleanup associated with the visible test-accuracy transition.
ΠΡΠΎΠ΄ΠΎΠ»ΠΆΠ°ΠΉΡΠ΅ ΡΡΠΈΡΡΡΡ
ΠΡΡΠ³ΠΈΠ΅ ΡΡΠΊΠΎΠ²ΠΎΠ΄ΡΡΠ²Π°, Π²ΡΠ±ΡΠ°Π½Π½ΡΠ΅ ΠΏΠΎ ΡΡΠΎΠΉ ΡΠ΅ΠΌΠ΅
ΠΠ°Π»Π΅Π΅Π‘Π»Π΅Π΄ΡΡΡΠ΅Π΅ ΡΡΠΊΠΎΠ²ΠΎΠ΄ΡΡΠ²ΠΎ
Π Π΅Π³ΡΠ»ΡΡΠΈΠ·Π°ΡΠΈΡ
ΠΡΠ½ΠΎΠ²Ρ