arXiv 2309.140622024-01-12cs.CVcs.LG

FeCAM：在无样本持续学习中利用类别分布的异质性

FeCAM: Exploiting the Heterogeneity of Class Distributions in Exemplar-Free Continual Learning

Universitat Autònoma de Barcelona（巴塞罗那自治大学）；Computer Vision Center, Barcelona University of Chinese Academy of Sciences（巴塞罗那大学计算机视觉中心，中国科学院）；State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences（中国科学院沈阳自动化研究所机器人学国家重点实验室）；Institutes for Robotics and Intelligent Manufacturing, Chinese Academy of Sciences（中国科学院机器人与智能制造研究所）

Dipam Goswami, Yuyang Liu, Bartłomiej Twardowski, Joost van de Weijer

备注：: Accepted at NeurIPS 2023

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出FeCAM方法，通过利用冻结特征提取器下的原型网络和各向异性马氏距离建模特征协方差关系，解决无样本类增量学习中特征分布异质性问题，在多样本、少样本及域增量场景下取得最先进结果。

摘要：

无样本类增量学习（CIL）带来了若干挑战，因为它禁止对先前任务的数据进行重放，因此会遭受灾难性遗忘。近年来，在第一个任务之后冻结特征提取器并增量学习分类器的方法受到了广泛关注。在本文中，我们探索了用于CIL的原型网络，该网络使用冻结的特征提取器生成新的类别原型，并基于特征到原型的欧氏距离进行分类。在对类别特征分布的分析中，我们表明基于欧氏度量的分类对于联合训练的特征是成功的。然而，当从非平稳数据中学习时，我们观察到欧氏度量是次优的，并且特征分布是异质的。为了应对这一挑战，我们重新审视了用于CIL的各向异性马氏距离。此外，我们通过实验表明，对特征协方差关系进行建模优于先前从正态分布中采样特征并训练线性分类器的尝试。与现有方法不同，我们的方法可推广到多样本和少样本CIL设置，以及域增量设置。有趣的是，在不更新骨干网络的情况下，我们的方法在多个标准持续学习基准上取得了最先进的结果。代码可在 https://github.com/dipamgoswami/FeCAM 获取。

英文摘要：

Exemplar-free class-incremental learning (CIL) poses several challenges since it prohibits the rehearsal of data from previous tasks and thus suffers from catastrophic forgetting. Recent approaches to incrementally learning the classifier by freezing the feature extractor after the first task have gained much attention. In this paper, we explore prototypical networks for CIL, which generate new class prototypes using the frozen feature extractor and classify the features based on the Euclidean distance to the prototypes. In an analysis of the feature distributions of classes, we show that classification based on Euclidean metrics is successful for jointly trained features. However, when learning from non-stationary data, we observe that the Euclidean metric is suboptimal and that feature distributions are heterogeneous. To address this challenge, we revisit the anisotropic Mahalanobis distance for CIL. In addition, we empirically show that modeling the feature covariance relations is better than previous attempts at sampling features from normal distributions and training a linear classifier. Unlike existing methods, our approach generalizes to both many- and few-shot CIL settings, as well as to domain-incremental settings. Interestingly, without updating the backbone network, our method obtains state-of-the-art results on several standard continual learning benchmarks. Code is available at https://github.com/dipamgoswami/FeCAM.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。