Scholarlink

<p>The parallelization of optimization algorithms is of paramount importance in large-scale machine learning. In this paper, we explore the implementation of Adaptive learning rate Stochastic Gradient Descent (A-SGD) in a synchronized and parallelized manner. Additionally, we incorporate a Variance Reduction (VR) strategy to enhance the rate of convergence. Our approach addresses the complexity associated with high-dimensional datasets, particularly within the context of Logistic Regression (LR) and Support Vector Machine (SVM). Initially, we utilize the Histogram of Oriented Gradients (HOG) to extract high-dimensional sparse features from a dataset designed for Blindness Detection. Subsequently, we employ LR and SVM as our classifiers of choice. Finally, we apply the Synchronous A-SGD (SA-SGD) and Synchronous Adaptive Stochastic Variance Reduction (SA-SVRG) to the solutions of these classifiers. Our experimental results indicate that the performance of SA-SGD and SA-SVRG is notably superior when executed on a cluster as opposed to a single node.</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

Large-scale machine learning with synchronous parallel adaptive stochastic variance reduction gradient descent for high-dimensional blindness detection on spark

  • 期刊论文

  • 发表日期：

    2025年3月9日

查看全文
  • Chuandong Qin,
  • Yiqing Zhang,
  • Yu Cao

摘要

The parallelization of optimization algorithms is of paramount importance in large-scale machine learning. In this paper, we explore the implementation of Adaptive learning rate Stochastic Gradient Descent (A-SGD) in a synchronized and parallelized manner. Additionally, we incorporate a Variance Reduction (VR) strategy to enhance the rate of convergence. Our approach addresses the complexity associated with high-dimensional datasets, particularly within the context of Logistic Regression (LR) and Support Vector Machine (SVM). Initially, we utilize the Histogram of Oriented Gradients (HOG) to extract high-dimensional sparse features from a dataset designed for Blindness Detection. Subsequently, we employ LR and SVM as our classifiers of choice. Finally, we apply the Synchronous A-SGD (SA-SGD) and Synchronous Adaptive Stochastic Variance Reduction (SA-SVRG) to the solutions of these classifiers. Our experimental results indicate that the performance of SA-SGD and SA-SVRG is notably superior when executed on a cluster as opposed to a single node.