We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Discovering Parametric Activation Functions

Abstract: Recent studies have shown that the choice of activation function can significantly affect the performance of deep learning networks. However, the benefits of novel activation functions have been inconsistent and task-dependent, and therefore the rectified linear unit (ReLU) is still the most commonly used. This paper proposes a technique for customizing activation functions automatically, resulting in reliable improvements in performance. Evolutionary search is used to discover the general form of the function, and gradient descent to optimize its parameters for different parts of the network and over the learning process. Experiments with three different neural network architectures on the CIFAR-100 image classification dataset show that this approach is effective. It discovers different activation functions for different architectures, and consistently improves accuracy over ReLU and other recently proposed activation functions by significant margins. The approach can therefore be used as an automated optimization step in applying deep learning to new tasks.
Comments: 11 pages, 6 figures/tables, under review
Subjects: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Neural and Evolutionary Computing (cs.NE); Machine Learning (stat.ML)
Cite as: arXiv:2006.03179 [cs.LG]
  (or arXiv:2006.03179v1 [cs.LG] for this version)

Submission history

From: Garrett Bingham [view email]
[v1] Fri, 5 Jun 2020 00:25:33 GMT (643kb,D)
[v2] Tue, 6 Oct 2020 15:33:14 GMT (3178kb,D)
[v3] Tue, 8 Dec 2020 19:28:47 GMT (3181kb,D)
[v4] Sat, 30 Jan 2021 02:17:20 GMT (3174kb,D)
[v5] Fri, 21 Jan 2022 19:39:36 GMT (8971kb,D)

Link back to: arXiv, form interface, contact.