What question did this study set out to answer?

To develop new metrics and a cost function to enhance training of Morphological-Linear Neural Networks.

April 24, 2026Open Access

IoU-based overlapping metric and aggregated loss function for morphological-linear neural networks

Key Points

To develop new metrics and a cost function to enhance training of Morphological-Linear Neural Networks.
Introduced the Overlapping Intersection over Union (OIoU) metric for quantifying neuron overlap.
Developed OIoU-Loss to optimize layer configurations during neural network training.
Implemented both contributions in morphological-linear neural networks.
OIoU achieved layer size reductions greater than 30%.
OIoU-Loss improved classification accuracy while maintaining geometric separation of decision regions.

Abstract

This article introduces two novel metrics based on Intersection over Union (IoU) for Morphological-Linear Neural Networks (MLNNs). The first contribution is the Overlapping Intersection over Union (OIoU) metric, which quantifies overlap among dendritic neurons in the first MLNN layer. This metric enables layer size reductions greater than 30%. The second contribution is OIoU-Loss, a cost function that integrates the OIoU metric into the training process. OIoU-Loss minimizes hyperbox overlap during training, achieving effective classification while preserving geometric separation between decision regions.

IoU-based overlapping metric and aggregated loss function for morphological-linear neural networks

Key Points

Abstract

Cite This Study