Evaluating the Performance of Explanation Methods on Ordinal Regression CNN Models

Research areas:

• Uncategorized

Year:

2023

Type of Publication:

In Proceedings

Keywords:

Convolutional Neural Networks, Interpretability, Ordinal Regression

Authors:

• Barbero-Gómez, Javier
• Cruz, Ricardo
• Cardoso, Jaime S.
• Gutiérrez, Pedro Antonio
• Hervás-Martínez, César

Volume:

14135

Book title:

IWANN 2023: Advances in Computational Intelligence

Series:

Lecture Notes in Computer Science (LNCS)

Pages:

529-540

Organization:

Ponta Delgada, Portugal

Month:

19th-21th June, 2023

BibTex:

@conference{EvaluatingPerformance_ExplanationMethods_2023, author = "Javier Barbero-G{\'o}mez and Ricardo Cruz and Jaime S. Cardoso and Pedro Antonio Guti{\'e}rrez and C{\'e}sar Herv{\'a}s-Mart{\'i}nez", abstract = "This paper introduces an evaluation procedure to validate the efficacy of explanation methods for Convolutional Neural Network (CNN) models in ordinal regression tasks. Two ordinal methods are contrasted against a baseline using cross-entropy, across four datasets. A statistical analysis demonstrates that attribution methods, such as Grad-CAM and IBA, perform significantly better when used with ordinal regression CNN models compared to a baseline approach in most ordinal and nominal metrics. The study suggests that incorporating ordinal information into the attribution map construction process may improve the explanations further.", booktitle = "IWANN 2023: Advances in Computational Intelligence", doi = "10.1007/978-3-031-43078-7_43", keywords = "Convolutional Neural Networks, Interpretability, Ordinal Regression", month = "19th-21th June, 2023", organization = "Ponta Delgada, Portugal", pages = "529--540", publisher = "Springer", series = "Lecture Notes in Computer Science (LNCS)", title = "{E}valuating the {P}erformance of {E}xplanation {M}ethods on {O}rdinal {R}egression {CNN} {M}odels", url = "link.springer.com/chapter/10.1007/978-3-031-43078-7_43", volume = "14135", year = "2023", }

Abstract:

This paper introduces an evaluation procedure to validate the efficacy of explanation methods for Convolutional Neural Network (CNN) models in ordinal regression tasks. Two ordinal methods are contrasted against a baseline using cross-entropy, across four datasets. A statistical analysis demonstrates that attribution methods, such as Grad-CAM and IBA, perform significantly better when used with ordinal regression CNN models compared to a baseline approach in most ordinal and nominal metrics. The study suggests that incorporating ordinal information into the attribution map construction process may improve the explanations further.