Label-Similarity Curriculum Learning

Publikation: Bidrag til bog/antologi/rapportKonferencebidrag i proceedingsForskningfagfællebedømt

Dokumenter

  • Fulltext

    Indsendt manuskript, 1,59 MB, PDF-dokument

Curriculum learning can improve neural network training by guiding the optimization to desirable optima. We propose a novel curriculum learning approach for image classification that adapts the loss function by changing the label representation. The idea is to use a probability distribution over classes as target label, where the class probabilities reflect the similarity to the true class. Gradually, this label representation is shifted towards the standard one-hot-encoding. That is, in the beginning minor mistakes are corrected less than large mistakes, resembling a teaching process in which broad concepts are explained first before subtle differences are taught. The class similarity can be based on prior knowledge. For the special case of the labels being natural words, we propose a generic way to automatically compute the similarities. The natural words are embedded into Euclidean space using a standard word embedding. The probability of each class is then a function of the cosine similarity between the vector representations of the class and the true label. The proposed label-similarity curriculum learning (LCL) approach was empirically evaluated using several popular deep learning architectures for image classification tasks applied to five datasets including ImageNet, CIFAR100, and AWA2. In all scenarios, LCL was able to improve the classification accuracy on the test data compared to standard training. Code to reproduce results is available at https://github.com/speedystream/LCL.

OriginalsprogEngelsk
TitelComputer Vision – ECCV 2020 - 16th European Conference, Proceedings
RedaktørerAndrea Vedaldi, Horst Bischof, Thomas Brox, Jan-Michael Frahm
Antal sider17
ForlagSpringer VS
Publikationsdato2020
Sider174-190
ISBN (Trykt)9783030585259
DOI
StatusUdgivet - 2020
Begivenhed16th European Conference on Computer Vision, ECCV 2020 - Glasgow, Storbritannien
Varighed: 23 aug. 202028 aug. 2020

Konference

Konference16th European Conference on Computer Vision, ECCV 2020
LandStorbritannien
ByGlasgow
Periode23/08/202028/08/2020
NavnLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Vol/bind12374 LNCS
ISSN0302-9743

ID: 250554869