The Slingshot Mechanism: An Empirical Study of Adaptive Optimizers and the Grokking Phenomenon
This work addresses a puzzling training behavior in machine learning that affects researchers and practitioners using adaptive optimizers, though it is incremental as it builds on prior grokking studies.
The paper investigates the grokking phenomenon, where models suddenly generalize after long overfitting, by identifying the Slingshot Mechanism, an optimization anomaly in adaptive optimizers that causes cyclic phase transitions and is linked to grokking onset, with empirical observations showing grokking occurs almost exclusively at slingshot onset without explicit regularization.
The grokking phenomenon as reported by Power et al. ( arXiv:2201.02177 ) refers to a regime where a long period of overfitting is followed by a seemingly sudden transition to perfect generalization. In this paper, we attempt to reveal the underpinnings of Grokking via a series of empirical studies. Specifically, we uncover an optimization anomaly plaguing adaptive optimizers at extremely late stages of training, referred to as the Slingshot Mechanism. A prominent artifact of the Slingshot Mechanism can be measured by the cyclic phase transitions between stable and unstable training regimes, and can be easily monitored by the cyclic behavior of the norm of the last layers weights. We empirically observe that without explicit regularization, Grokking as reported in ( arXiv:2201.02177 ) almost exclusively happens at the onset of Slingshots, and is absent without it. While common and easily reproduced in more general settings, the Slingshot Mechanism does not follow from any known optimization theories that we are aware of, and can be easily overlooked without an in depth examination. Our work points to a surprising and useful inductive bias of adaptive gradient optimizers at late stages of training, calling for a revised theoretical analysis of their origin.