Scholarlink

<p>To address the challenges associated with the fragmented nature of healthcare resulting in centralized data sharing and the violation of data privacy, federated learning offers a cutting-edge paradigm that enables decentralized clients to cooperatively train machine learning models while keeping the data localized. This research proposes a Federated Learning-based framework called DenseFedCNN framework having a novel DenseNet201-based CNN model called DenseConNet, coupled with Client Specific Hyperparameter Tuning (CSHT) algorithm for multi-class classification of Brain MRI images. CSHT helps in dealing with the adaptability of this model with the clients’ data distribution and in faster convergence with better accuracy. The work considers both the scenarios of independent and identically distributed (IID) data and Non-independent and identically distributed (Non-IID) data. The challenge of NonIIDScenario has been tackled by ensuring the quantity skewness among the client data distribution. For the effectiveness of using CSHT along with the DenseConNet model, experiments are performed on both the IIDScenario and NonIIDScenario with and without CSHT. The performance of the DenseConNet model with CSHT in both the IID and non-IID scenarios is observed to be approximately equal to the model with centralized data-sharing coupled with excellent values of the truth loss resulting in an efficient classification framework without violating data privacy.</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

Densefedcnn: a federated learning-based framework with client-specific hyperparameter tuning for multi-class brain MRI classification

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Naima Firdaus,
  • Zahid Raza

摘要

To address the challenges associated with the fragmented nature of healthcare resulting in centralized data sharing and the violation of data privacy, federated learning offers a cutting-edge paradigm that enables decentralized clients to cooperatively train machine learning models while keeping the data localized. This research proposes a Federated Learning-based framework called DenseFedCNN framework having a novel DenseNet201-based CNN model called DenseConNet, coupled with Client Specific Hyperparameter Tuning (CSHT) algorithm for multi-class classification of Brain MRI images. CSHT helps in dealing with the adaptability of this model with the clients’ data distribution and in faster convergence with better accuracy. The work considers both the scenarios of independent and identically distributed (IID) data and Non-independent and identically distributed (Non-IID) data. The challenge of NonIIDScenario has been tackled by ensuring the quantity skewness among the client data distribution. For the effectiveness of using CSHT along with the DenseConNet model, experiments are performed on both the IIDScenario and NonIIDScenario with and without CSHT. The performance of the DenseConNet model with CSHT in both the IID and non-IID scenarios is observed to be approximately equal to the model with centralized data-sharing coupled with excellent values of the truth loss resulting in an efficient classification framework without violating data privacy.