Regularizing with Pseudo-Negatives for Continual Self-Supervised Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cha, Sungmin, Cho, Kyunghyun, Moon, Taesup
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917687104372736
author Cha, Sungmin
Cho, Kyunghyun
Moon, Taesup
author_facet Cha, Sungmin
Cho, Kyunghyun
Moon, Taesup
contents We introduce a novel Pseudo-Negative Regularization (PNR) framework for effective continual self-supervised learning (CSSL). Our PNR leverages pseudo-negatives obtained through model-based augmentation in a way that newly learned representations may not contradict what has been learned in the past. Specifically, for the InfoNCE-based contrastive learning methods, we define symmetric pseudo-negatives obtained from current and previous models and use them in both main and regularization loss terms. Furthermore, we extend this idea to non-contrastive learning methods which do not inherently rely on negatives. For these methods, a pseudo-negative is defined as the output from the previous model for a differently augmented version of the anchor sample and is asymmetrically applied to the regularization term. Extensive experimental results demonstrate that our PNR framework achieves state-of-the-art performance in representation learning during CSSL by effectively balancing the trade-off between plasticity and stability.
format Preprint
id arxiv_https___arxiv_org_abs_2306_05101
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Regularizing with Pseudo-Negatives for Continual Self-Supervised Learning
Cha, Sungmin
Cho, Kyunghyun
Moon, Taesup
Machine Learning
We introduce a novel Pseudo-Negative Regularization (PNR) framework for effective continual self-supervised learning (CSSL). Our PNR leverages pseudo-negatives obtained through model-based augmentation in a way that newly learned representations may not contradict what has been learned in the past. Specifically, for the InfoNCE-based contrastive learning methods, we define symmetric pseudo-negatives obtained from current and previous models and use them in both main and regularization loss terms. Furthermore, we extend this idea to non-contrastive learning methods which do not inherently rely on negatives. For these methods, a pseudo-negative is defined as the output from the previous model for a differently augmented version of the anchor sample and is asymmetrically applied to the regularization term. Extensive experimental results demonstrate that our PNR framework achieves state-of-the-art performance in representation learning during CSSL by effectively balancing the trade-off between plasticity and stability.
title Regularizing with Pseudo-Negatives for Continual Self-Supervised Learning
topic Machine Learning
url https://arxiv.org/abs/2306.05101