An Exploration of Softmax Alternatives Belonging to the Spherical Loss Family

TL;DR

This paper investigates alternative loss functions within the spherical family as substitutes for softmax in multi-class classification, revealing promising results on certain datasets despite some limitations.

Contribution

It introduces and evaluates spherical loss functions, including a new log-Taylor Softmax, expanding the options beyond traditional softmax for neural network training.

Findings

Spherical loss functions outperform softmax on MNIST and CIFAR-10.

Log-softmax remains superior on language modeling tasks.

New log-Taylor Softmax shows potential as an alternative.

Abstract

In a multi-class classification problem, it is standard to model the output of a neural network as a categorical distribution conditioned on the inputs. The output must therefore be positive and sum to one, which is traditionally enforced by a softmax. This probabilistic mapping allows to use the maximum likelihood principle, which leads to the well-known log-softmax loss. However the choice of the softmax function seems somehow arbitrary as there are many other possible normalizing functions. It is thus unclear why the log-softmax loss would perform better than other loss alternatives. In particular Vincent et al. (2015) recently introduced a class of loss functions, called the spherical family, for which there exists an efficient algorithm to compute the updates of the output weights irrespective of the output size. In this paper, we explore several loss functions from this family as possible alternatives to the traditional log-softmax. In particular, we focus our investigation on spherical bounds of the log-softmax loss and on two spherical log-likelihood losses, namely the log-Spherical Softmax suggested by Vincent et al. (2015) and the log-Taylor Softmax that we introduce. Although these alternatives do not yield as good results as the log-softmax loss on two language modeling tasks, they surprisingly outperform it in our experiments on MNIST and CIFAR-10, suggesting that they might be relevant in a broad range of applications.