Evaluation of Neural Architectures Trained with Square Loss vs Cross-Entropy in Classification Tasks

Abstract

Modern neural architectures for classification tasks are trained using the cross-entropy loss, which is widely believed to be empirically superior to the square loss. In this work we provide evidence indicating that this belief may not be well-founded. We explore several major neural architectures and a range of standard benchmark datasets for NLP, automatic speech recognition (ASR) and computer vision tasks to show that these architectures, with the same hyper-parameter settings as reported in the literature, perform comparably or better when trained with the square loss, even after equalizing computational resources. Indeed, we observe that the square loss produces better results in the dominant majority of NLP and ASR experiments. Cross-entropy appears to have a slight edge on computer vision tasks. We argue that there is little compelling empirical or theoretical evidence indicating a clear-cut advantage to the cross-entropy loss. Indeed, in our experiments, performance on nearly all non-vision tasks can be improved, sometimes significantly, by switching to the square loss. Furthermore, training with square loss appears to be less sensitive to the randomness in initialization. We posit that training using the square loss for classification needs to be a part of best practices of modern deep learning on equal footing with cross-entropy.

An extended version of the paper published at ICLR2021. Added material includes evaluations of Transformer architectures

References in corpus (2)

• Dive into Deep Learning
• On Loss Functions for Deep Neural Networks in Classification

Cited by in corpus (14)

• Gradient Starvation: A Learning Proclivity in Neural Networks
• Neural Collapse Under MSE Loss: Proximity to and Dynamics on the Central Path
• Benign Overfitting in Multiclass Classification: All Roads Lead to Interpolation
• A Farewell to the Bias-Variance Tradeoff? An Overview of the Theory of Overparameterized Machine Learning
• On the interplay between data structure and loss function in classification problems
• An Effective, Robust and Fairness-aware Hate Speech Detection Framework
• Exploring the Uncertainty Properties of Neural Networks' Implicit Priors in the Infinite-Width Limit
• RATT: Leveraging Unlabeled Data to Guarantee Generalization
• Estimating informativeness of samples with Smooth Unique Information
• Selective Regression Under Fairness Criteria
• Memorization in Deep Neural Networks: Does the Loss Function matter?
• On the Regularization of Autoencoders
• On Sampling-Based Training Criteria for Neural Language Modeling
• Prediction of Hereditary Cancers Using Neural Networks