This project uses Kolmogorov-Arnold Networks (KANs) within Surrogate-assisted Evolutionary Algorithms (SAEAs) to efficiently solve black-box optimization problems by reducing the dependency on function evaluations. The code supports the use of KANs for regression and classification, improving optimization efficiency through model assisted selection, thus curtailing function calls. The experiments prove the model's effective integration, showcasing enhanced performance in optimization tasks.
cd KAN-EA
conda create -n KANEA python=3.10
conda activate KANEA
pip install -r requirements.txt
python setup.py develop
python run_kan_sps.py
run KAN-SPS on Ellipsoid function
from pymoo.optimize import minimize
from problem.LZG import LZG01
from model.KAN_surrogate import KAN_Regressor,KAN_Classifier
from alg.pre_selection import KAN_SPS
problem = LZG01(n_var=5)
model = KAN_Regressor() # USE KAN_Regressor or KAN_Classifier
# model = KAN_Classifier()
algorithm = KAN_SPS(problem=problem,
pop_size=50,
n_offsprings=50,
reproduction_type="CoDE",
trial_vectors_num=3,
model=model
)
res = minimize(problem,
algorithm,
('n_evals', 2000),
verbose=True)run KAN-SAS on Ellipsoid function
from problem.LZG import LZG01
from pymoo.optimize import minimize
from alg.ueda import KAN_SAS
if __name__=='__main__':
problem = LZG01(n_var=5)
algorithm = KAN_SAS(pop_size=50)
res = minimize(problem,
algorithm,
('n_evals', 300),
verbose=True)The current training process of the KANs model is quite time-consuming and has poor compatibility on CUDA. For efficiency, you can refer to some efficient implementations in awesome-kan project.
@misc{hao2024look,
title={A First Look at Kolmogorov-Arnold Networks in Surrogate-assisted Evolutionary Algorithms},
author={Hao Hao and Xiaoqun Zhang and Bingdong Li and Aimin Zhou},
year={2024},
eprint={2405.16494},
archivePrefix={arXiv},
primaryClass={cs.NE}
}