Continuous Hyper-parameter Configuration for Particle Swarm Optimization via Auto-tuning
| Autor: | Vladimir Milián Núñez, Jairo Rojas-Delgado, Rafael Bello, Rafael A. Trujillo-Rasúa |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Hyperparameter
021110 strategic defence & security studies Mathematical optimization Optimization problem Optimization algorithm Computer science 0211 other engineering and technologies Particle swarm optimization 02 engineering and technology Field (computer science) Auto tuning Random search Hyperparameter optimization 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Hardware_ARITHMETICANDLOGICSTRUCTURES |
| Zdroj: | Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications ISBN: 9783030339036 CIARP |
| DOI: | 10.1007/978-3-030-33904-3_43 |
| Popis: | Hyper-Parameter configuration is a relatively novel field of paramount importance in machine learning and optimization. Hyper-parameters refers to the parameters that control the behavior of algorithms and are not tuned directly by such algorithms. For hyper-parameters of an optimization algorithm such as Particle Swarm Optimization, hyper-parameter configuration is a nested optimization problem. Usually, practitioners needs to use a second optimization algorithm such as grid search or random search to find proper hyper-parameters. However, this approach forces practitioners to know about two different algorithms. Moreover, hyper-parameter configuration algorithms also have hyper-parameters that need to be considered. In this work we use Particle Swarm Optimization to configure its own hyper-parameters. Results show that hyper-parameters configured by PSO are competitive with hyper-parameters found by other hyper-parameter configuration algorithms. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|