Enhancing The Explainability of Neural Network: In the aspects of feature importance, domain rules and algorithmic transparency - Brossura

Arulprakash, Enoch; Aruldoss, Martin

 
9786208064341: Enhancing The Explainability of Neural Network: In the aspects of feature importance, domain rules and algorithmic transparency
Brossura
ISBN 10: 6208064341 ISBN 13: 9786208064341
Casa editrice: LAP LAMBERT Academic Publishing, 2024

Sinossi

Artificial Intelligence (AI) driven by neural networks is crucial in many applications like recommendation systems, language translation, social media, chatbots, and spell-checking etc. However, these networks are often criticized for being "black boxes," raising concerns about their explainability, especially in sensitive domains like healthcare, autonomous driving etc. Existing methods to enhance explainability, such as feature importance, often lack clarity and interpretability.To address this, the Object-Oriented Neural Network for Improved Explainability (OONNIE) was developed. OONNIE uses object-oriented modeling to combine loss and connection weight for computing feature importance and integrates domain-specific rules through OOP's extendability. The model emphasizes algorithmic transparency by detailing every training step. Evaluated on XOR and XNOR functions, OONNIE shows promising results in feature importance, faster loss reduction, and improved predictions after integrating domain rules. This marks a significant contribution to explainable AI, making OONNIE a valuable tool for developing trustworthy AI systems.

Le informazioni nella sezione "Riassunto" possono far riferimento a edizioni diverse di questo titolo.

Editore
LAP LAMBERT Academic Publishing
Data di pubblicazione
2024
Lingua
Inglese
ISBN 10 
6208064341
ISBN 13 
9786208064341
Rilegatura
Copertina flessibile
Numero di pagine
204
Contatto del produttore
non disponibile
Persona responsabile
KreGa Onlinehandel GmbH
4950000000000
info@krega-onlinehandel.de

Birkenweg 31
Brakel
33034
Germania