Saved in:
Bibliographic Details
Main Authors: Zhou, Zhangchen, Zhang, Yaoyu, Xu, Zhi-Qin John
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2405.17479
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916262680985600
author Zhou, Zhangchen
Zhang, Yaoyu
Xu, Zhi-Qin John
author_facet Zhou, Zhangchen
Zhang, Yaoyu
Xu, Zhi-Qin John
contents Grokking is the phenomenon where neural networks NNs initially fit the training data and later generalize to the test data during training. In this paper, we empirically provide a frequency perspective to explain the emergence of this phenomenon in NNs. The core insight is that the networks initially learn the less salient frequency components present in the test data. We observe this phenomenon across both synthetic and real datasets, offering a novel viewpoint for elucidating the grokking phenomenon by characterizing it through the lens of frequency dynamics during the training process. Our empirical frequency-based analysis sheds new light on understanding the grokking phenomenon and its underlying mechanisms.
format Preprint
id arxiv_https___arxiv_org_abs_2405_17479
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle A rationale from frequency perspective for grokking in training neural network
Zhou, Zhangchen
Zhang, Yaoyu
Xu, Zhi-Qin John
Machine Learning
Neural and Evolutionary Computing
Grokking is the phenomenon where neural networks NNs initially fit the training data and later generalize to the test data during training. In this paper, we empirically provide a frequency perspective to explain the emergence of this phenomenon in NNs. The core insight is that the networks initially learn the less salient frequency components present in the test data. We observe this phenomenon across both synthetic and real datasets, offering a novel viewpoint for elucidating the grokking phenomenon by characterizing it through the lens of frequency dynamics during the training process. Our empirical frequency-based analysis sheds new light on understanding the grokking phenomenon and its underlying mechanisms.
title A rationale from frequency perspective for grokking in training neural network
topic Machine Learning
Neural and Evolutionary Computing
url https://arxiv.org/abs/2405.17479