Marginal Replay vs Conditional Replay for Continual Learning

  • Timothée Lesort
  • , Alexander Gepperth
  • , Andrei Stoian
  • , David Filliat

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

Abstract

We present a new replay-based method of continual classification learning that we term “conditional replay” which generates samples and labels together by sampling from a distribution conditioned on the class. We compare conditional replay to another replay-based continual learning paradigm (which we term “marginal replay”) that generates samples independently of their class and assigns labels in a separate step. The main improvement in conditional replay is that labels for generated samples need not be inferred, which reduces the margin for error in complex continual classification learning tasks. We demonstrate the effectiveness of this approach using novel and standard benchmarks constructed from MNIST and FashionMNIST data, and compare to the regularization-based elastic weight consolidation (EWC) method [17, 34].

Original languageEnglish
Title of host publicationArtificial Neural Networks and Machine Learning – ICANN 2019
Subtitle of host publicationDeep Learning - 28th International Conference on Artificial Neural Networks, Proceedings
EditorsIgor V. Tetko, Pavel Karpov, Fabian Theis, Vera Kurková
PublisherSpringer Verlag
Pages466-480
Number of pages15
ISBN (Print)9783030304836
DOIs
Publication statusPublished - 1 Jan 2019
Event28th International Conference on Artificial Neural Networks: Workshop and Special Sessions, ICANN 2019 - Munich, Germany
Duration: 17 Sept 201919 Sept 2019

Publication series

NameLecture Notes in Computer Science
Volume11728 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference28th International Conference on Artificial Neural Networks: Workshop and Special Sessions, ICANN 2019
Country/TerritoryGermany
CityMunich
Period17/09/1919/09/19

Keywords

  • Continual learning
  • Generative models
  • Generative replay

Fingerprint

Dive into the research topics of 'Marginal Replay vs Conditional Replay for Continual Learning'. Together they form a unique fingerprint.

Cite this