A fuzzy loss for ontology classification

TL;DR

This paper introduces a fuzzy loss function for ontology classification that enhances logical consistency of deep learning models by penalizing violations of ontology constraints, without sacrificing classification accuracy.

Contribution

The paper presents a novel fuzzy loss that integrates logical constraints into deep learning, significantly reducing ontology violations in classification tasks.

Findings

Fuzzy loss reduces consistency violations by several orders of magnitude.

The approach maintains classification performance while improving logical consistency.

Unsupervised learning with fuzzy loss further enhances ontology compliance.

Abstract

Deep learning models are often unaware of the inherent constraints of the task they are applied to. However, many downstream tasks require logical consistency. For ontology classification tasks, such constraints include subsumption and disjointness relations between classes. In order to increase the consistency of deep learning models, we propose a fuzzy loss that combines label-based loss with terms penalising subsumption- or disjointness-violations. Our evaluation on the ChEBI ontology shows that the fuzzy loss is able to decrease the number of consistency violations by several orders of magnitude without decreasing the classification performance. In addition, we use the fuzzy loss for unsupervised learning. We show that this can further improve consistency on data from a