Toward Automated Modeling of Abstract Concepts and Natural Phenomena: Autoencoding Straight Lines

Yuval Bayer, David Harel, Assaf Marron, Smadar Szekely

2023

Abstract

Modeling complex systems or natural phenomena requires special skills and extensive domain knowledge. This makes automating model development an intriguing challenge. One question is whether a model’s ontology—the essence of its entities—can be learned automatically from observation. We describe work in progress on automating the learning of a basic concept: an image of the straight line segment between two points in a two-dimensional plane. Humans readily encode such images using two endpoints, or a point, an angle, and a length. Furthermore, image recognition algorithms readily detect line segments in images. Here, we employ autoencoders. Autoencoders perform both feature extraction and reconstruction of inputs from their coded representation. It turns out that autoencoding line segments is not trivial. Our interim conclusions include: (1) Developing methods for comparing the performance of different autoencoders in a given task is an essential research challenge. (2) Development of autoencoders manifests supervision of this purportedly unsupervised process; one then asks what knowledge employed in such development can be obtained automatically. (3) Automatic modeling of properties of observed objects requires multiple representations and sensors. This work can eventually benefit broader issues in automated model development.

Paper Citation

in Harvard Style

Bayer Y., Harel D., Marron A. and Szekely S. (2023). Toward Automated Modeling of Abstract Concepts and Natural Phenomena: Autoencoding Straight Lines. In Proceedings of the 11th International Conference on Model-Based Software and Systems Engineering - Volume 1: MODELSWARD, ISBN 978-989-758-633-0, pages 275-282. DOI: 10.5220/0011886100003402

in Bibtex Style

@conference{modelsward23,
author={Yuval Bayer and David Harel and Assaf Marron and Smadar Szekely},
title={Toward Automated Modeling of Abstract Concepts and Natural Phenomena: Autoencoding Straight Lines},
booktitle={Proceedings of the 11th International Conference on Model-Based Software and Systems Engineering - Volume 1: MODELSWARD,},
year={2023},
pages={275-282},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0011886100003402},
isbn={978-989-758-633-0},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 11th International Conference on Model-Based Software and Systems Engineering - Volume 1: MODELSWARD,
TI - Toward Automated Modeling of Abstract Concepts and Natural Phenomena: Autoencoding Straight Lines
SN - 978-989-758-633-0
AU - Bayer Y.
AU - Harel D.
AU - Marron A.
AU - Szekely S.
PY - 2023
SP - 275
EP - 282
DO - 10.5220/0011886100003402