.png){kind=icon}
.png){kind=icon}
Return
A new parallel bat colony optimization algorithm and its application for evolving CNN architectures with artificial selection
.png){kind=icon}
PRE
AI
DOI：10.1007/s00500-023-08846-x.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Bat Algorithm is an evolutionary metaheuristic that is used to find solutions to NP-Hard problems like the Travelling Salesman Problem. A Parallel Bat Colony Optimization Algorithm (PBCOA) based on the introduction of colonies to the popular Bat Algorithm is proposed in this article. The concept of colonies is introduced in the Bat Algorithm along with phenomena like reproductive crossover and migration. The resultant Parallel Bat Colony Optimization Algorithm is tested on 30 standard benchmark Functions, on which it shows clear superiority over 17 other optimization algorithms. PBCOA is applied to evolve CNN architectures over landmark recognition datasets. The learning rate and layers in the network were artificially selected as 0.9 and 3 respectively to quicken the process of training and make the model lesser complex. Even under the specified time and memory constraints, PBCOA improves the network accuracy and F1 Score from around 0.80 and less to more than 0.99. The evolved network, when tested on larger datasets, also shows an unprecedented improvement in performance, thereby meaning that the evolution of CNNs with PBCOA can be highly desirable.
Keywords:
Bat algorithm
Evolutionary metaheuristics
Benchmarking
Convolutional neural networks
Image recognition