Optimized Particle Swarm Optimization for Neural Network Training
Author Information
Author(s): Michael Meissner, Michael Schmuker, Gisbert Schneider
Primary Institution: Johann Wolfgang Goethe-Universität, Institut für Organische Chemie und Chemische Biologie
Hypothesis
Can the performance of Particle Swarm Optimization (PSO) be improved through meta-optimization of its parameters?
Conclusion
Optimizing the free parameters of the PSO method can result in performance gain.
Supporting Evidence
- The OPSO method improved optimization speed and quality in most experiments.
- Training time decreased by a factor of four compared to other PSO methods.
- The OPSO approach yielded a prediction model with good correlation with training, test, and validation data.
Takeaway
This study shows that by adjusting the settings of a particle swarm algorithm, we can make it work better and faster, especially for training neural networks.
Methodology
The study used a meta-optimization approach to improve PSO parameters and tested it on various artificial fitness functions and neural network training.
Limitations
The study primarily focused on specific test functions and may not generalize to all optimization problems.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website