Home // COGNITIVE 2019, The Eleventh International Conference on Advanced Cognitive Technologies and Applications // View article

Few-Shot Learning Using Supervised Non-Associative Autoencoders and Correlation Techniques

Authors:
Ehsan Sedgh Gooya
Ayman Al Falou

Keywords: Networks; Few-shot Supervised learning; Autoencoders

Abstract:
Deep learning, while very effective today, traditionally requires very large amounts of labeled data to perform the classification task. In an attempt to solve this problem, the few-shot learning concept, which uses few labeled samples by class, becomes more and more useful. In this paper, we propose a new low-shot learning method, dubbed Supervised Non-Associative Auto-Encoder (SNAAE) to perform classification. Complementary to prior studies, SNAAE represents a shift of paradigm in comparison with the usual few-shot learning methods, as it does not use any prior knowledge neither unlabeled data. SNAAE is based on stacking layers of an autoencoder, which are trained in a supervised way to rebuild a single version representing their inputs. The reconstructed output is then classified outside of the neural network by correlation plane quantification metric. To perform the classification, the rebuilt output is compared with the initial versions used as target to train the SNAAE. We demonstrate empirically the efficiency of our proposed approach on the well known handwritten digits Modified National Institute of Standards and Technology database (MNIST) database.

Pages: 1 to 2

Copyright: Copyright (c) IARIA, 2019

Publication date: May 5, 2019

Published in: conference

ISSN: 2308-4197

ISBN: 978-1-61208-705-4

Location: Venice, Italy

Dates: from May 5, 2019 to May 9, 2019