Global sensitivity estimates for neural network classifiers

Áreas de investigación:

• Methodology - Machine learning

Año:

2017

Tipo de publicación:

Artículo

Autores:

• Fernandez-Navarro, Francisco
• Carbonero-Ruz, Mariano
• Becerra-Alonso, David
• Torres-Jiménez, Mercedes

Journal:

IEEE Transactions on Neural Networks and Learning Systems

Volumen:

28

Número:

11

Páginas:

2592-2604

ISSN:

2162-237X

BibTex:

@article{122016, author = "Francisco Fernandez-Navarro and Mariano Carbonero-Ruz and David Becerra-Alonso and Mercedes Torres-Jim{\'e}nez", abstract = "Artificial Neural Networks (ANNs) have traditionally been seen as black-box models because, although they are able to find “hidden” relations between inputs and outputs with a high approximation capacity, their structure seldom provides any insights on the structure of the functions being approximated. Several research papers have tried to debunk the black-box nature of ANNs, since it limits the potential use of ANNs in many research areas. The present article is framed in this context and proposes a methodology to determine the individual and collective effects of the input variables on the outputs for classification problems based on the ANOVA-functional decomposition. The method is applied after the training phase of the ANN and allows researchers to rank the input variables according to their importance in the variance of the ANN output. The computation of the sensitivity indices for Product Unit Neural Networks (PUNNs) is straightforward as those indices can be calculated analytically by evaluating the integrals in the ANOVA decomposition. Unfortunately, the sensitivity indices associated to ANNs based on sigmoidal basis functions or radial basis functions can not be calculated analytically. In this manuscript, the indices for those kind of ANNs are proposed to be estimated by the (quasi) Monte Carlo method.", awards = "JCR(2017): 7.982 Position: 6/133 (Q1) Category: COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE", comments = "JCR(2017): 7.982 Position: 6/133 (Q1) Category: COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE", doi = "10.1109/TNNLS.2016.2598657", issn = "2162-237X ", journal = "IEEE Transactions on Neural Networks and Learning Systems", note = "JCR(2017): 7.982 Position: 6/133 (Q1) Category: COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE", number = "11", pages = "2592-2604", title = "{G}lobal sensitivity estimates for neural network classifiers", url = "http://dx.doi.org/10.1109/TNNLS.2016.2598657", volume = "28", year = "2017", }

Nota:

JCR(2017): 7.982 Position: 6/133 (Q1) Category: COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Abstract:

Artificial Neural Networks (ANNs) have traditionally been seen as black-box models because, although they are able to find “hidden” relations between inputs and outputs with a high approximation capacity, their structure seldom provides any insights on the structure of the functions being approximated. Several research papers have tried to debunk the black-box nature of ANNs, since it limits the potential use of ANNs in many research areas. The present article is framed in this context and proposes a methodology to determine the individual and collective effects of the input variables on the outputs for classification problems based on the ANOVA-functional decomposition. The method is applied after the training phase of the ANN and allows researchers to rank the input variables according to their importance in the variance of the ANN output. The computation of the sensitivity indices for Product Unit Neural Networks (PUNNs) is straightforward as those indices can be calculated analytically by evaluating the integrals in the ANOVA decomposition. Unfortunately, the sensitivity indices associated to ANNs based on sigmoidal basis functions or radial basis functions can not be calculated analytically. In this manuscript, the indices for those kind of ANNs are proposed to be estimated by the (quasi) Monte Carlo method.

Comentarios:

JCR(2017): 7.982 Position: 6/133 (Q1) Category: COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE