Genetic algorithm for automatic tuning of neural network hyperparameters

Jakub Safarik; Jakub Jalowiczor; Erik Gresak; Jan Rozhon

doi:10.1117/12.2304955

3 May 2018 Genetic algorithm for automatic tuning of neural network hyperparameters

Jakub Safarik, Jakub Jalowiczor, Erik Gresak, Jan Rozhon

Proceedings Volume 10643, Autonomous Systems: Sensors, Vehicles, Security, and the Internet of Everything; 106430Q (2018) https://doi.org/10.1117/12.2304955
Event: SPIE Defense + Security, 2018, Orlando, FL, United States

Abstract

Artificial neural networks affect our everyday life. But every neural network depends on the appropriate training set and setting of internal properties with hyperparameters. Even accurate and complete training set doesnt imply high performance of neural network algorithm. Tuning of hyperparameters is crucial for correct functionality, fast learning and high accuracy of neural networks. The hyperparameter selection relies on manual fine-tuning based on multiple full training trials. There are a lot of neural network implementation available for public and commercial use, but the setting of hyperparameters is often a neglected problem. Choosing the best structure and hyperparameters is the primary challenge in designing a neural network. This article describes a genetic algorithm for automatic selection of hyperparameters and their tuning for increasing the performance of neural networks without human interaction. The optimization algorithm accelerates the discovery of configuration, which is otherwise a time-consuming task. We evaluate the results of optimizations in comparison to naïve approach and compare pro and cons of different techniques.

Citation Download Citation

Jakub Safarik, Jakub Jalowiczor, Erik Gresak, and Jan Rozhon "Genetic algorithm for automatic tuning of neural network hyperparameters", Proc. SPIE 10643, Autonomous Systems: Sensors, Vehicles, Security, and the Internet of Everything, 106430Q (3 May 2018); https://doi.org/10.1117/12.2304955

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available