Apprentissage de relations floues pour l’annotation sémantique expliquée avec peu de données

Abstract : Despite the recent successes of deep learning, such models are still far from some human abilities like learning from few examples, reasoning and explaining their decisions. In this paper, we focus on object annotation in images and we introduce a reasoning framework that is based on learning fuzzy relations on a dataset for generating explanations. Given a catalogue of relations, it efficiently induces the most relevant relations and combines them for building constraints in order to both solve the target task and generate explanations. We test our approach on a publicly available dataset on which the goal was both to perform multiple organ annotation and to provide explanations. We show that our model can generate explanations and achieve high performance despite being trained on a small dataset containing as few as a couple of examples.
Complete list of metadatas

Cited literature [39 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02160290
Contributor : Florent Breuil <>
Submitted on : Thursday, June 20, 2019 - 5:14:11 PM
Last modification on : Wednesday, September 4, 2019 - 1:40:15 PM

File

rjcia2019-03.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02160290, version 1

Citation

Régis Pierrard, Jean-Philippe Poli, Céline Hudelot. Apprentissage de relations floues pour l’annotation sémantique expliquée avec peu de données. Rencontres des Jeunes Chercheurs en Intelligence Artificielle 2019, Toulouse Institute of Computer Science Research (IRIT), and the French AI Society (AFIA), Jul 2019, Toulouse, France. pp.18-26. ⟨hal-02160290⟩

Share

Metrics

Record views

175

Files downloads

171