Scholarlink

This study presents a novel model that utilizes Multiple Kernel Learning (MKL) within Generative Adversarial Networks (GANs) through Robust Optimization (RO) techniques. Additionally, Stochastic Functional Gradient (SFG) is incorporated alongside MKL. A thorough comparison was made between this new model and traditional approaches such as Empirical Risk Minimization (ERM), sample average approximation, and the SFG Reproducing Kernel Hilbert Spaces (SFG-RKHS). The latter integrates RKHS, which are fundamental to Support Vector Machines (SVM), with SFG. Several challenging datasets, including CIFAR 10, CIFAR 100, MNIST, Fashion MNIST, EMNIST, and SVHN are considered for the evaluation. Our results demonstrate that the SFG MKL model is a strong candidate for machine learning applications that demand accuracy and robustness against adversarial influences. Exploring various kernels within the MKL framework opens new avenues for advancement, particularly in critical sectors like biomedical imaging and autonomous systems. Merging gradients with MKL presents a promising opportunity to elevate the standards of reliable machine-learning algorithms.

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

A Novel Robust Optimization Model: The Synergy of Stochastic Functional Gradients and Kernel Methods in Machine Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mohammed Thamer Kamil Al-Khazraji,
  • Duygu Üçüncü,
  • Süreyya Akyüz

摘要

This study presents a novel model that utilizes Multiple Kernel Learning (MKL) within Generative Adversarial Networks (GANs) through Robust Optimization (RO) techniques. Additionally, Stochastic Functional Gradient (SFG) is incorporated alongside MKL. A thorough comparison was made between this new model and traditional approaches such as Empirical Risk Minimization (ERM), sample average approximation, and the SFG Reproducing Kernel Hilbert Spaces (SFG-RKHS). The latter integrates RKHS, which are fundamental to Support Vector Machines (SVM), with SFG. Several challenging datasets, including CIFAR 10, CIFAR 100, MNIST, Fashion MNIST, EMNIST, and SVHN are considered for the evaluation. Our results demonstrate that the SFG MKL model is a strong candidate for machine learning applications that demand accuracy and robustness against adversarial influences. Exploring various kernels within the MKL framework opens new avenues for advancement, particularly in critical sectors like biomedical imaging and autonomous systems. Merging gradients with MKL presents a promising opportunity to elevate the standards of reliable machine-learning algorithms.