Transformation-Equivariant Representation Learning with Barber-Agakov and InfoNCE Mutual Information Estimation

Marshal Sinaga, T. Basarrudin, Adila Krisnadhi

2022

Abstract

The success of deep learning on computer vision tasks is due to the convolution layer being equivariant to the translation. Several works attempt to extend the notion of equivariance into more general transformations. Autoencoding variational transformation (AVT) achieves state of art by approaching the problem from the information theory perspective. The model involves the computation of mutual information, which leads to a more general transformation-equivariant representation model. In this research, we investigate the alternatives of AVT called variational transformation-equivariant (VTE). We utilize the Barber-Agakov and information noise contrastive mutual information estimation to optimize VTE. Furthermore, we also propose a sequential mechanism that involves a self-supervised learning model called predictive-transformation to train our VTE. Results of experiments demonstrate that VTE outperforms AVT on image classification tasks.

Paper Citation

in Harvard Style

Sinaga M., Basarrudin T. and Krisnadhi A. (2022). Transformation-Equivariant Representation Learning with Barber-Agakov and InfoNCE Mutual Information Estimation. In Proceedings of the 11th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM, ISBN 978-989-758-549-4, pages 99-109. DOI: 10.5220/0010880400003122

in Bibtex Style

@conference{icpram22,
author={Marshal Sinaga and T. Basarrudin and Adila Krisnadhi},
title={Transformation-Equivariant Representation Learning with Barber-Agakov and InfoNCE Mutual Information Estimation},
booktitle={Proceedings of the 11th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM,},
year={2022},
pages={99-109},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0010880400003122},
isbn={978-989-758-549-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 11th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM,
TI - Transformation-Equivariant Representation Learning with Barber-Agakov and InfoNCE Mutual Information Estimation
SN - 978-989-758-549-4
AU - Sinaga M.
AU - Basarrudin T.
AU - Krisnadhi A.
PY - 2022
SP - 99
EP - 109
DO - 10.5220/0010880400003122