The opaqueness of deep neural networks hinders their employment in safety-critical applications. This issue is driving the research community to focus on eXplainable Artificial Intelligence (XAI) techniques. XAI algorithms can be categorized into two types: those that explain the predictions of black-box models and those that create interpretable models from the start. While interpretable models foster user trust, their performance is normally inferior to traditional black-box models like neural networks. To fill this gap, this chapter presents an extensive framework introducing special neural networks known as Logic Explained Networks (LENs). The most notable advantage of this approach is that LENs achieve performances that are comparable to the state-of-the-art neural networks while providing concise First-Order Logic explanations. Moreover, LENs are designed to learn various types of explanations, such as local or global, by associating related task functions or directly explaining a single class based on the others and/or the input data. They can offer explanations for their behaviour or that of other black-box models. Finally, LENs can be applied in many learning scenarios: they can be trained using supervised or unsupervised learning and applied across different domains like tabular data, computer vision, natural language processing, and relational tasks.

Learning logic explanations by neural networks / Ciravegna, G.; Giannini, F.; Barbiero, P.; Gori, M.; Lio, P.; Maggini, M.; Melacci, S.. - (2023), pp. 547-558. - FRONTIERS IN ARTIFICIAL INTELLIGENCE AND APPLICATIONS. [10.3233/FAIA230157].

Learning logic explanations by neural networks

Lio P.
;
2023

Abstract

The opaqueness of deep neural networks hinders their employment in safety-critical applications. This issue is driving the research community to focus on eXplainable Artificial Intelligence (XAI) techniques. XAI algorithms can be categorized into two types: those that explain the predictions of black-box models and those that create interpretable models from the start. While interpretable models foster user trust, their performance is normally inferior to traditional black-box models like neural networks. To fill this gap, this chapter presents an extensive framework introducing special neural networks known as Logic Explained Networks (LENs). The most notable advantage of this approach is that LENs achieve performances that are comparable to the state-of-the-art neural networks while providing concise First-Order Logic explanations. Moreover, LENs are designed to learn various types of explanations, such as local or global, by associating related task functions or directly explaining a single class based on the others and/or the input data. They can offer explanations for their behaviour or that of other black-box models. Finally, LENs can be applied in many learning scenarios: they can be trained using supervised or unsupervised learning and applied across different domains like tabular data, computer vision, natural language processing, and relational tasks.
2023
Compendium of Neurosymbolic Artificial Intelligence
9781643684062
9781643684079
.
02 Pubblicazione su volume::02a Capitolo o Articolo
Learning logic explanations by neural networks / Ciravegna, G.; Giannini, F.; Barbiero, P.; Gori, M.; Lio, P.; Maggini, M.; Melacci, S.. - (2023), pp. 547-558. - FRONTIERS IN ARTIFICIAL INTELLIGENCE AND APPLICATIONS. [10.3233/FAIA230157].
File allegati a questo prodotto
Non ci sono file associati a questo prodotto.

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: https://hdl.handle.net/11573/1725190
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact