Learning Combinations of Activation Functions

TL;DR

This paper proposes two methods to automatically learn and optimize combinations of activation functions during neural network training, leading to improved performance on standard datasets.

Contribution

It introduces novel approaches for learning activation function combinations dynamically during training, moving beyond fixed, manually tuned non-linearities.

Findings

Improved top-1 accuracy on ImageNet by 3.01 percentage points for AlexNet.

Significant performance gains on Fashion-MNIST and CIFAR-10 datasets.

Demonstrated effectiveness across multiple well-known architectures.

Abstract

In the last decade, an active area of research has been devoted to design novel activation functions that are able to help deep neural networks to converge, obtaining better performance. The training procedure of these architectures usually involves optimization of the weights of their layers only, while non-linearities are generally pre-specified and their (possible) parameters are usually considered as hyper-parameters to be tuned manually. In this paper, we introduce two approaches to automatically learn different combinations of base activation functions (such as the identity function, ReLU, and tanh) during the training phase. We present a thorough comparison of our novel approaches with well-known architectures (such as LeNet-5, AlexNet, and ResNet-56) on three standard datasets (Fashion-MNIST, CIFAR-10, and ILSVRC-2012), showing substantial improvements in the overall performance, such as an increase in the top-1 accuracy for AlexNet on ILSVRC-2012 of 3.01 percentage points.