Aligned Contrastive Loss for Long-Tailed Recognition

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ma, Jiali, Cui, Jiequan, Kazuki, Maeno, Subramanian, Lakshmi, Jayashree, Karlekar, Pranata, Sugiri, Zhang, Hanwang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912408948178944
author Ma, Jiali
Cui, Jiequan
Kazuki, Maeno
Subramanian, Lakshmi
Jayashree, Karlekar
Pranata, Sugiri
Zhang, Hanwang
author_facet Ma, Jiali
Cui, Jiequan
Kazuki, Maeno
Subramanian, Lakshmi
Jayashree, Karlekar
Pranata, Sugiri
Zhang, Hanwang
contents In this paper, we propose an Aligned Contrastive Learning (ACL) algorithm to address the long-tailed recognition problem. Our findings indicate that while multi-view training boosts the performance, contrastive learning does not consistently enhance model generalization as the number of views increases. Through theoretical gradient analysis of supervised contrastive learning (SCL), we identify gradient conflicts, and imbalanced attraction and repulsion gradients between positive and negative pairs as the underlying issues. Our ACL algorithm is designed to eliminate these problems and demonstrates strong performance across multiple benchmarks. We validate the effectiveness of ACL through experiments on long-tailed CIFAR, ImageNet, Places, and iNaturalist datasets. Results show that ACL achieves new state-of-the-art performance.
format Preprint
id arxiv_https___arxiv_org_abs_2506_01071
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Aligned Contrastive Loss for Long-Tailed Recognition
Ma, Jiali
Cui, Jiequan
Kazuki, Maeno
Subramanian, Lakshmi
Jayashree, Karlekar
Pranata, Sugiri
Zhang, Hanwang
Computer Vision and Pattern Recognition
In this paper, we propose an Aligned Contrastive Learning (ACL) algorithm to address the long-tailed recognition problem. Our findings indicate that while multi-view training boosts the performance, contrastive learning does not consistently enhance model generalization as the number of views increases. Through theoretical gradient analysis of supervised contrastive learning (SCL), we identify gradient conflicts, and imbalanced attraction and repulsion gradients between positive and negative pairs as the underlying issues. Our ACL algorithm is designed to eliminate these problems and demonstrates strong performance across multiple benchmarks. We validate the effectiveness of ACL through experiments on long-tailed CIFAR, ImageNet, Places, and iNaturalist datasets. Results show that ACL achieves new state-of-the-art performance.
title Aligned Contrastive Loss for Long-Tailed Recognition
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.01071