Scholarlink

Federated Learning (FL) is Machine Learning (ML) algorithm which performs the collaborative and distribute learning when private information present locally on client. To ensure the privacy of data, gradients are deformed or their representation is disturbed before sharing, that minimizes the performance of method. Recently, existing methods has drawbacks like problem of gradient leakage of FL in cloud that undermines data privacy. In this research, the Swarm Learning (SL) based Federated Learning (FL) in cloud environment is proposed to enhance the data privacy in cloud. In SL based FL, instead of sharing the disturbed or gradients with noise to central server, here shares the actual gradients between authenticate training nodes. The performance of proposed SL based FL method is evaluated with metrics of Time To Deploy (TTD), Time To Complete (TTC), utilization of Central Processing Unit (CPU) and Random Access Memory (RAM) on various cloud nodes (C1, C2, C3, C4). The proposed SL based FL method attained the less TTD of 27s, TTC of 4.18h and CPU Utilization of 0.34% on C1, 1.15% on C2, 1.64% on C3 and 1.98% on C4 which is effective than other techniques like Fuzzy Analytical Hierarchy Process (FAHP).

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

Swarm Learning Based Federated Learning for Data Privacy in Cloud Environment

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Madhavi Najana,
  • Arun Pandiyan Perumal,
  • Pradeep Chintale,
  • Laxminarayana Korada,
  • Ankur Mahida,
  • Sri Harsha Vardhan Sanne

摘要

Federated Learning (FL) is Machine Learning (ML) algorithm which performs the collaborative and distribute learning when private information present locally on client. To ensure the privacy of data, gradients are deformed or their representation is disturbed before sharing, that minimizes the performance of method. Recently, existing methods has drawbacks like problem of gradient leakage of FL in cloud that undermines data privacy. In this research, the Swarm Learning (SL) based Federated Learning (FL) in cloud environment is proposed to enhance the data privacy in cloud. In SL based FL, instead of sharing the disturbed or gradients with noise to central server, here shares the actual gradients between authenticate training nodes. The performance of proposed SL based FL method is evaluated with metrics of Time To Deploy (TTD), Time To Complete (TTC), utilization of Central Processing Unit (CPU) and Random Access Memory (RAM) on various cloud nodes (C1, C2, C3, C4). The proposed SL based FL method attained the less TTD of 27s, TTC of 4.18h and CPU Utilization of 0.34% on C1, 1.15% on C2, 1.64% on C3 and 1.98% on C4 which is effective than other techniques like Fuzzy Analytical Hierarchy Process (FAHP).