Class-Incremental Learning with Rectified Feature-Graph Preservation

Cheng Hsun Lei, Yi Hsin Chen, Wen Hsiao Peng, Wei Chen Chiu*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

In this paper, we address the problem of distillation-based class-incremental learning with a single head. A central theme of this task is to learn new classes that arrive in sequential phases over time while keeping the model’s capability of recognizing seen classes with only limited memory for preserving seen data samples. Many regularization strategies have been proposed to mitigate the phenomenon of catastrophic forgetting. To understand better the essence of these regularizations, we introduce a feature-graph preservation perspective. Insights into their merits and faults motivate our weighted-Euclidean regularization for old knowledge preservation. We further propose rectified cosine normalization and show how it can work with binary cross-entropy to increase class separation for effective learning of new classes. Experimental results on both CIFAR-100 and ImageNet datasets demonstrate that our method outperforms the state-of-the-art approaches in reducing classification error, easing catastrophic forgetting, and encouraging evenly balanced accuracy over different classes. Our project page is at : https://github.com/yhchen12101/FGP-ICL.

Original languageEnglish
Title of host publicationComputer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers
EditorsHiroshi Ishikawa, Cheng-Lin Liu, Tomas Pajdla, Jianbo Shi
PublisherSpringer Science and Business Media Deutschland GmbH
Pages358-374
Number of pages17
ISBN (Print)9783030695439
DOIs
StatePublished - 2021
Event15th Asian Conference on Computer Vision, ACCV 2020 - Virtual, Online
Duration: 30 Nov 20204 Dec 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12627 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference15th Asian Conference on Computer Vision, ACCV 2020
CityVirtual, Online
Period30/11/204/12/20

Fingerprint Dive into the research topics of 'Class-Incremental Learning with Rectified Feature-Graph Preservation'. Together they form a unique fingerprint.

Cite this