Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets
It addresses the poorly understood problem of generalization in overparametrized neural networks for researchers in deep learning, though it is incremental in nature.
The paper investigates neural network generalization on small algorithmic datasets, finding that networks can achieve perfect generalization through a 'grokking' process that occurs after overfitting, with performance improving from random chance to perfect levels.
In this paper we propose to study generalization of neural networks on small algorithmically generated datasets. In this setting, questions about data efficiency, memorization, generalization, and speed of learning can be studied in great detail. In some situations we show that neural networks learn through a process of "grokking" a pattern in the data, improving generalization performance from random chance level to perfect generalization, and that this improvement in generalization can happen well past the point of overfitting. We also study generalization as a function of dataset size and find that smaller datasets require increasing amounts of optimization for generalization. We argue that these datasets provide a fertile ground for studying a poorly understood aspect of deep learning: generalization of overparametrized neural networks beyond memorization of the finite training dataset.