Scholarlink

<p>Training artificial neural networks is often challenged by large search spaces, slow convergence, and premature stagnation in local optima. To address these challenges, this work introduces an Enhanced Bat Algorithm (EBA) that incorporates multiple improvement strategies to enhance search robustness and ensure stable convergence. The proposed EBA is applied to optimize Gaussian centers within radial basis function (RBF) networks. Beyond optimizing Gaussian centers, we introduce a joint RBF neural-network training mechanism that simultaneously determines RBF widths and synaptic weights through a new optimization technique fulfilled by Optimum Weight Vector (OWV), the Adam optimizer, and <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\:{L}_{2}\)</EquationSource> </InlineEquation> regularization to improve generalization. Evaluation on 30 benchmark optimization functions in very high dimensional demonstrates that EBA delivers the best performance, achieves the lowest average mean error across all categories, and significantly outperforms well-known optimization algorithms including BA, PSO, GA, ACO, ALO, CSO, FO, and GWO—often by several orders of magnitude. In image recognition tasks, the proposed RBF + EBA + Adam + OWV classifier achieves state-of-the-art performance on four datasets: 97.1% (QT), 89.6% (SUN-397), 96.7% (COCO), and 82.6% (NUS-WIDE). These results represent improvements of 8.4%, 12.1%, 11.6%, and 15.9% over standard RBF networks and consistently outperform conventional SVM, MLP, and deep feature-fusion approaches. The findings confirm that the proposed joint-optimization framework is a powerful and robust solution for both high-dimensional optimization and large-scale image classification.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Joint optimization of radial basis function neural networks using an enhanced bat algorithm and adaptive weight tuning for robust image recognition

  • 期刊论文

  • 发表日期：

    2026年5月19日

查看全文
  • Davar Giveki

摘要

Training artificial neural networks is often challenged by large search spaces, slow convergence, and premature stagnation in local optima. To address these challenges, this work introduces an Enhanced Bat Algorithm (EBA) that incorporates multiple improvement strategies to enhance search robustness and ensure stable convergence. The proposed EBA is applied to optimize Gaussian centers within radial basis function (RBF) networks. Beyond optimizing Gaussian centers, we introduce a joint RBF neural-network training mechanism that simultaneously determines RBF widths and synaptic weights through a new optimization technique fulfilled by Optimum Weight Vector (OWV), the Adam optimizer, and \(\:{L}_{2}\) regularization to improve generalization. Evaluation on 30 benchmark optimization functions in very high dimensional demonstrates that EBA delivers the best performance, achieves the lowest average mean error across all categories, and significantly outperforms well-known optimization algorithms including BA, PSO, GA, ACO, ALO, CSO, FO, and GWO—often by several orders of magnitude. In image recognition tasks, the proposed RBF + EBA + Adam + OWV classifier achieves state-of-the-art performance on four datasets: 97.1% (QT), 89.6% (SUN-397), 96.7% (COCO), and 82.6% (NUS-WIDE). These results represent improvements of 8.4%, 12.1%, 11.6%, and 15.9% over standard RBF networks and consistently outperform conventional SVM, MLP, and deep feature-fusion approaches. The findings confirm that the proposed joint-optimization framework is a powerful and robust solution for both high-dimensional optimization and large-scale image classification.