Scholarlink

<p>As cloud-based financial systems become more interconnected, the fight against sophisticated frauds requires privacy-preserving adaptive intelligence. The given paper introduces an agentic federated cybersecurity approach that combines Spiking Neural Networks (SNNs) with Algae optimization algorithm (AOA) to develop self-adaptive fraud detection agents in cloud-based financial systems. In contrast to centralized systems, the proposed framework applies Federated Learning (FL) to train local models from private transaction data without revealing raw inputs. The nodes operate independently, learning temporal fraud patterns using SNNs and adjusting detection behavior with algae-inspired metaheuristic algorithm. The agents adaptively change spike thresholds, learning rates, and feature selection to stay responsive to changing fraud environments. The proposed framework enhances fraud detection performance by 30%, decreases false positives by 25%, and decreases detection latency by 20% in contrast to the standard centralized models. By integrating agentic learning, federated intelligence, and nature-inspired optimization, the model provides an expandable, secure, and context-aware solution for fraud avoidance in contemporary financial networks.</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

FedSNA: a federated learning neuro-spiking and algae-optimized agentic AI framework for real-time fraud detection in cloud-based financial services

  • 期刊论文

  • 发表日期：

    2026年1月10日

查看全文
  • Rajesh Sura,
  • Mohan Kumar Meesala,
  • Prayas Lohalekar,
  • Nitya Sri Nellore,
  • Nitin Mukhi,
  • Mahesh Kumar Goyal

摘要

As cloud-based financial systems become more interconnected, the fight against sophisticated frauds requires privacy-preserving adaptive intelligence. The given paper introduces an agentic federated cybersecurity approach that combines Spiking Neural Networks (SNNs) with Algae optimization algorithm (AOA) to develop self-adaptive fraud detection agents in cloud-based financial systems. In contrast to centralized systems, the proposed framework applies Federated Learning (FL) to train local models from private transaction data without revealing raw inputs. The nodes operate independently, learning temporal fraud patterns using SNNs and adjusting detection behavior with algae-inspired metaheuristic algorithm. The agents adaptively change spike thresholds, learning rates, and feature selection to stay responsive to changing fraud environments. The proposed framework enhances fraud detection performance by 30%, decreases false positives by 25%, and decreases detection latency by 20% in contrast to the standard centralized models. By integrating agentic learning, federated intelligence, and nature-inspired optimization, the model provides an expandable, secure, and context-aware solution for fraud avoidance in contemporary financial networks.