logo
科言猫学术研究的AI总结
language
手机端手机端
  • Phone Feature微信小程序
简体中文简体中文
notice消息通知
博客博客
delete
arrow
返回

Schedule-Robust Continual Learning

delete2025-10-03
delete0
delete
OA
AI
R
Ruohan Wang
M
Marco Ciccone
M
Massimiliano Pontil
C
Carlo Ciliberto
DOI：10.1109/TPAMI.2025.3614868delete
delete原文链接
delete分享
delete收藏
查看原文
摘要

摘要

En 中文
持续学习（CL）解决机器学习中的一个基本挑战，旨在从非平稳数据流中持续学习新数据，同时缓解先前学习数据的遗忘问题。尽管现有的CL算法已引入多种应对遗忘的实用技术，但鲜有关注研究数据调度（即决定数据流样本分布随时间演化的方式）如何影响CL问题。经验上，大多数CL方法易受调度变化的影响：当处理同一底层训练数据上的“更困难”调度时，其准确率显著降低。在实际场景中，数据调度往往是未知的，因此关键挑战在于设计能够抵抗多样化调度的CL方法，以确保模型可靠性。本文引入了CL的新概念——**schedule robustness**，并提出Schedule-Robust Continual Learning（SCROLL），一种满足此期望属性的强大基线方法。SCROLL在适当地预训练表示上训练线性分类器，随后仅使用回放数据进行模型适应。我们将SCROLL与CL的元学习表述相联系，并证明了其在调度鲁棒性方面的保证。经验上，所提出的方法显著优于现有的CL方法，并提供详尽的消融实验以突出其特性。
Keyword:
Continual learning
lifelong learning
meta-learning
representation learning
AI总结

AI总结

对已上传原文的论文进行重点信息的提取，主要内容包括：简要概述、研究摘要、背景介绍、关键亮点、图文解析、展望与总结。

期刊

IEEE Transactions on Pattern Analysis and Machine Intelligence 封面图
IEEE Transactions on Pattern Analysis and Machine Intelligence
IF:
18.6
论文数:
1.1K
被引数:
9.8W

机构

V
vector institute
学者数:
9
论文数: 5
被引数: 0
A
agency for science, technology, and research
学者数:
48
论文数: 20
被引数: 0
U
university college london
学者数:
8.9K
论文数: 4.8K
被引数: 1
学者 查看更多机构
引用论文

引用论文

Learning to Prompt for Continual Learning
err2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
IF0
err2022-06-01
err0
errOAAI
errZifeng Wang; Zizhao Zhang; Chen-Yu Lee; Han Zhang; Ruoxi Sun; Xiaoqi Ren; Guolong Su; Vincent Perot; Jennifer Dy; Tomas Pfister
err分享
err收藏
REMIND Your Neural Network to Prevent Catastrophic Forgetting
errLecture Notes in Computer Science
IF0
err2020-11-07
err0
PREAI
errTyler L. Hayes; Kushal Kafle; Robik Shrestha; Manoj Acharya; Christopher Kanan
err分享
err收藏
Automated Flower Classification over a Large Number of Classes在大量课程上进行自动花卉分类
err2008 Sixth Indian Conference on Computer Vision, Graphics & Image Processing
IF0
err2008-12-01
err0
PREAI
errMaria-Elena Nilsback; Andrew Zisserman
err分享
err收藏
SS-IL: Separated Softmax for Incremental LearningSs-il: 用于增量学习的分离Softmax
err2021 IEEE/CVF International Conference on Computer Vision (ICCV)
IF0
err2021-10-01
err0
errOAAI
errHongjoon Ahn; Jihwan Kwak; Subin Lim; Hyeonsu Bang; Hyojun Kim; Taesup Moon
err分享
err收藏
Linear Discriminant Analysis线性判别分析
errSpringerBriefs in Optimization
IF0
err2012-11-02
err0
PREAI
errPetros Xanthopoulos; Panos M. Pardalos; Theodore B. Trafalis
err分享
err收藏
Task-Free Continual Learning
err2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
IF0
err2019-06-01
err0
errOAAI
errRahaf Aljundi; Klaas Kelchtermans; Tinne Tuytelaars
err分享
err收藏
CODA-Prompt: COntinual Decomposed Attention-Based Prompting for Rehearsal-Free Continual Learning
err2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
IF0
err2023-06-01
err0
errOAAI
errJames Seale Smith; Leonid Karlinsky; Vyshnavi Gutta; Paola Cascante-Bonilla; Donghyun Kim; Assaf Arbelle; Rameswar Panda; Rogerio Feris; Zsolt Kira
err分享
err收藏
Robust Meta-Representation Learning via Global Label Inference and Classification通过全局标签推理和分类的鲁棒元表示学习
errIEEE Transactions on Pattern Analysis and Machine Intelligence
IF0
err2024-04-01
err0
errOAAI
errRuohan Wang; John Isak Texas Falk; Massimiliano Pontil; Carlo Ciliberto
err分享
err收藏
ImageNet Large Scale Visual Recognition ChallengeImageNet大规模视觉识别挑战
errINTERNATIONAL JOURNAL OF COMPUTER VISION
IF9.3
err2015-04-11
err2.7W
PREAI
errRussakovsky, Olga; Deng, Jia; Su, Hao; Krause, Jonathan; Satheesh, Sanjeev; Ma, Sean; Huang, Zhiheng; Karpathy, Andrej; Khosla, Aditya; Bernstein, Michael; Berg, Alexander C.; Fei-Fei, Li
err分享
err收藏
A continual learning survey: Defying forgetting in classification tasks持续学习调查: 在分类任务中无视遗忘
errIEEE Transactions on Pattern Analysis and Machine Intelligence
IF0
err2021-01-01
err0
errOAAI
errMatthias Delange; Rahaf Aljundi; Marc Masana; Sarah Parisot; Xu Jia; Ales Leonardis; Greg Slabaugh; Tinne Tuytelaars
err分享
err收藏
学者 查看更多内容