Trainable Highly-expressive Activation Functions

TL;DR

This paper introduces DiTAC, a trainable, highly-expressive activation function based on diffeomorphic transformations, which improves neural network performance across various tasks with minimal additional parameters.

Contribution

The paper proposes DiTAC, a novel trainable activation function leveraging diffeomorphic transformations, enhancing expressiveness and outperforming existing functions in multiple deep learning tasks.

Findings

DiTAC outperforms fixed and other trainable activation functions in various tasks.

DiTAC achieves substantial performance improvements with few trainable parameters.

DiTAC is effective in semantic segmentation, image generation, regression, and classification.

Abstract

Nonlinear activation functions are pivotal to the success of deep neural nets, and choosing the appropriate activation function can significantly affect their performance. Most networks use fixed activation functions (e.g., ReLU, GELU, etc.), and this choice might limit their expressiveness. Furthermore, different layers may benefit from diverse activation functions. Consequently, there has been a growing interest in trainable activation functions. In this paper, we introduce DiTAC, a trainable highly-expressive activation function based on an efficient diffeomorphic transformation (called CPAB). Despite introducing only a negligible number of trainable parameters, DiTAC enhances model expressiveness and performance, often yielding substantial improvements. It also outperforms existing activation functions (regardless whether the latter are fixed or trainable) in tasks such as semantic segmentation, image generation, regression problems, and image classification. Our code is available at https://github.com/BGU-CS-VIL/DiTAC.