Representation and Interpretability of IE Integral Neural Networks

Aoi Honda, Yudai Kamata, Simon James

Abstract: While there has been a lot of research attention given to neural networks and other black-box machine learning methods, recent works on aggregation functions and fuzzy sets have highlighted the appeal of incorporating fuzzy integrals into network implementations in order to achieve interpretability. We present an application of the recently proposed inclusion-exclusion integral neural network to the Boston House-Price dataset to illustrate its potential and examine the settings leading to better performance.