.png){kind=icon}
.png){kind=icon}
Return
A multi-strategy collaborative hybrid intelligent optimization algorithm (HPSR) and its application in deep network hyperparameter optimization
.png){kind=icon}
PRE
AI
C
L
X
DOI：10.1007/s10586-026-06390-5.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This research introduces a multi-strategy collaborative hybrid intelligent optimization algorithm (HPSR) specifically engineered to address the curse of dimensionality and local optima challenges in deep learning hyperparameter optimization. The proposed methodology extends the conventional particle swarm optimization framework through the systematic integration of four innovative mechanisms. A spiral update strategy enhances local exploitation capabilities, while a random-step mechanism effectively mitigates convergence to local optima. The architecture further incorporates an adaptive velocity adjustment strategy with dynamic convergence factors and implements a Gaussian Process Regression surrogate model enhanced by Expected Improvement criteria for efficient candidate solution evaluation. Comprehensive experimental evaluations are conducted across multiple neural network architectures. Standard benchmark image classification datasets are employed to rigorously assess algorithmic generalizability. The proposed methodology demonstrates superior convergence properties and optimization stability compared to established baseline approaches, particularly in scenarios requiring a sophisticated balance between exploration and exploitation phases. Empirical results indicate that the synergistic combination of complementary optimization strategies effectively mitigates common limitations associated with conventional swarm intelligence algorithms in hyperparameter search applications, yielding statistically significant performance improvements.
Keywords:
Swarm intelligence optimization
Hyperparameter optimization
Spiral update strategy
Random-step mechanism
Journal
C
IF:
4.1
Papers:
4.8K
Citations:
7.5K
Organization
Cited Papers