Companion Classification Losses for Regression Problems

EasyChair Preprint 15784

Abstract

By their very nature, regression problems can be transformed into classification problems by discretizing their target variable. Within this perspective, in this work we investigate the possibility of improving the performance of deep machine learning models in regression scenarios through a training strategy that combines different classification and regression objectives. In particular, we train deep neural networks using the mean squared error along with categorical cross-entropy and the novel Fisher loss as companion losses. Finally, we will compare experimentally the results of these companion loss methods with the ones obtained using the standard mean squared loss.

Keyphrases: Categorical cross-entropy, Deep Neural Networks, Fisher loss, Mean Squared Error, Representation Learning, companion losses

@booklet{EasyChair:15784,
  author    = {Aitor Sánchez-Ferrera and Jose R. Dorronsoro Ibero},
  title     = {Companion Classification Losses for Regression Problems},
  howpublished = {EasyChair Preprint 15784},
  year      = {EasyChair, 2025}}