Training neural networks end-to-end for hyperbox-based classification
Denis Mayr Lima Martins, Christian Lülf, Fabian Cristian Gieseke · Neurocomputing · 2024
Modern decision-making requires the use of powerful algorithms to make sense of a variety of data. In this context, hyperbox induction has been seen as a promising technique in which decisions on the data are represented as a series of orthogonal, multidimensional boxes (i.e., hyperboxes) that are often interpretable and human-readable. However, existing hyperbox induction methods are no longer capable of efficiently handling the increasing volumes of data many application domains are confronted with. Moreover, current methods offer little to no control on specific properties of the induced box models, such as the number or the sizes of the hyperboxes. In this work, we propose a novel, fully differentiable framework for hyperbox induction that makes use of recent advancement in neural networks. In contrast to existing approaches, our hyperbox-based models can be trained in an end-to-end fashion, which leads to significantly reduced training times and superior classification results.