Deep Learning models have recently achieved incredible performances in the Computer Vision field and are being deployed in an ever-growing range of real-life scenarios. Since they do not intrinsically provide insights of their inner decision processes, the field of eXplainable Artificial Intelligence emerged. Different XAI techniques have already been proposed, but the existing literature lacks methods to quantitatively compare different explanations, and in particular the semantic component is systematically overlooked. In this paper we introduce quantitative and ontology-based techniques and metrics in order to enrich and compare different explanations and XAI algorithms.

Quantitative and Ontology-Based Comparison of Explanations for Image Classification

Alan Perotti;Rossano Schifanella
Last
2019

Abstract

Deep Learning models have recently achieved incredible performances in the Computer Vision field and are being deployed in an ever-growing range of real-life scenarios. Since they do not intrinsically provide insights of their inner decision processes, the field of eXplainable Artificial Intelligence emerged. Different XAI techniques have already been proposed, but the existing literature lacks methods to quantitatively compare different explanations, and in particular the semantic component is systematically overlooked. In this paper we introduce quantitative and ontology-based techniques and metrics in order to enrich and compare different explanations and XAI algorithms.
5th Annual Conference on machine Learning, Optimization and Data science (LOD)
Certosa di Pontignano, Siena – Tuscany, Italy
September 10-13, 2019
Lecture Notes in Computer Science
Springer, Cham
11943
58
70
978-3-030-37598-0
978-3-030-37599-7
https://link.springer.com/chapter/10.1007/978-3-030-37599-7_6
Deep learning; Deep neural networks; Image classification; Machine learning; Neural networks; Ontology; Semantics
Valentina Ghidini, Alan Perotti, Rossano Schifanella
File in questo prodotto:
File Dimensione Formato  
Ghidini2019_Chapter_QuantitativeAndOntology-BasedC.pdf

Accesso riservato

Tipo di file: PDF EDITORIALE
Dimensione 2.02 MB
Formato Adobe PDF
2.02 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/2318/1795559
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? 0
social impact