Scholarlink

Performance and many different deep learning methods have attracted attention in recent years for their application in solving multi-class classification problems. This paper proposes a method to improve the accuracy and efficiency of multiclass classification problems by integrating multiple deep learning models into hybrid models. The research examines how well Transformer models, Recurrent Neural Networks (RNNs), and Convolutional Neural Networks (CNNs) work together to handle various difficult datasets. The proposed method takes advantage of the advantages of each architecture: Transformers are used to capture long-range relationships, CNNs are used for feature extraction, and RNNs are used for sequence learning. This approach aims to minimize the computational load and maximize the classification performance by combining these models. This research is evaluated using three significant datasets: CIFAR-10, CIFAR-100, and Tiny ImageNet. The hybrid model obtains a test accuracy of 90.0% on CIFAR-10, 72.0% on CIFAR-100, and 90.5% on Tiny ImageNet, showing significant generalization across different data sources. The results demonstrate the model's versatility and efficacy in managing complex, real-world data. This research advances the creation of intelligent systems that can analyze and handle complex data, enabling more advanced applications in areas such as autonomous systems, healthcare, and education.

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

Optimizing Hybrid CNN-RNN-Transformer Models for Multiclass Classification Tasks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mustafa Kemal Ambar

摘要

Performance and many different deep learning methods have attracted attention in recent years for their application in solving multi-class classification problems. This paper proposes a method to improve the accuracy and efficiency of multiclass classification problems by integrating multiple deep learning models into hybrid models. The research examines how well Transformer models, Recurrent Neural Networks (RNNs), and Convolutional Neural Networks (CNNs) work together to handle various difficult datasets. The proposed method takes advantage of the advantages of each architecture: Transformers are used to capture long-range relationships, CNNs are used for feature extraction, and RNNs are used for sequence learning. This approach aims to minimize the computational load and maximize the classification performance by combining these models. This research is evaluated using three significant datasets: CIFAR-10, CIFAR-100, and Tiny ImageNet. The hybrid model obtains a test accuracy of 90.0% on CIFAR-10, 72.0% on CIFAR-100, and 90.5% on Tiny ImageNet, showing significant generalization across different data sources. The results demonstrate the model's versatility and efficacy in managing complex, real-world data. This research advances the creation of intelligent systems that can analyze and handle complex data, enabling more advanced applications in areas such as autonomous systems, healthcare, and education.