Learning Stabilization of Inclusion-Exclusion Integral Neural Networks via Mutual Learning

Abstract

In this study, we introduce mutual learning into the Inclusive-Exclusive Integral Neural Network (IEINN), which uses Inclusion-Exclusive Integral based on fuzzy measures and t-norms and other aggregation functions. Mutual learning is a method in which multiple models learn simultaneously by utilizing each other's outputs, and it is expected to improve learning stability and generalization performance. In this study, we compared the baseline IEINN model with the IEINN model applying mutual learning across multiple regression tasks, verifying its effectiveness from perspectives such as learning curve behavior and consistency of feature importance. Although the number of experiments was limited, it was confirmed that mutual learning contributes to learning stabilization.