Scholarlink

<p>Cloud-native applications depend on efficient management of latency, as it ensures timely data processing and delivery, which is essential for both performance and the user experience. Traditional models on centralized decision-making applied to large-scale distributed systems usually suffer with scalability, resilience, and efficiency. In this paper, we propose a novel distributed multi-agent mechanism using Federated Swarm Learning (FSL) to address these challenges. This system allows agents to act independently, makes decisions based on local data, and cooperate to attain the global optimization. Using the Fire Hawk Learning, every single agent maintains a trade-off between local and global decision-making. The proposed system allows several edge devices to jointly train deep learning models using Federated Learning (FL), without sharing raw data. This reduces the bandwidth consumption and improves privacy. Swarm learning allows devices to communicate directly with one another unlike traditional FL’s. This increases system resilience and reduces the central point failure risk. The simulation results indicate that performance has improved, where the proposed FSL approach provides 15% increased fault tolerance ability, and lowers the latency rate between 20 and 25%, and hence increases the model accuracy by 18%. Unlike centralized models, the proposed method reduces the energy consumption by 10% and increases scalability by 30%. These results show an increase in resilience and performance of cloud-native applications in distributed decision-making.</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

A novel federated swarm learning for increased resilience in latency-aware cloud-native applications

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Pooja Batra Nagpal,
  • Sunil Kumar,
  • Ankit Khare,
  • Venkatadri Marriboyina,
  • Ambika Aggarwal,
  • Hammam Alshazly

摘要

Cloud-native applications depend on efficient management of latency, as it ensures timely data processing and delivery, which is essential for both performance and the user experience. Traditional models on centralized decision-making applied to large-scale distributed systems usually suffer with scalability, resilience, and efficiency. In this paper, we propose a novel distributed multi-agent mechanism using Federated Swarm Learning (FSL) to address these challenges. This system allows agents to act independently, makes decisions based on local data, and cooperate to attain the global optimization. Using the Fire Hawk Learning, every single agent maintains a trade-off between local and global decision-making. The proposed system allows several edge devices to jointly train deep learning models using Federated Learning (FL), without sharing raw data. This reduces the bandwidth consumption and improves privacy. Swarm learning allows devices to communicate directly with one another unlike traditional FL’s. This increases system resilience and reduces the central point failure risk. The simulation results indicate that performance has improved, where the proposed FSL approach provides 15% increased fault tolerance ability, and lowers the latency rate between 20 and 25%, and hence increases the model accuracy by 18%. Unlike centralized models, the proposed method reduces the energy consumption by 10% and increases scalability by 30%. These results show an increase in resilience and performance of cloud-native applications in distributed decision-making.