Scholarlink

Wireless Access Networks (WANs) and especially self-configurable systems in smart cities, the Army, and mass IoT ecosystems are becoming increasingly popular as a key to the seamless connectivity and automation. However, they are very dynamic and thus prone to zero-day faults - faults or failures never previously experienced, which arise in changing environments, with new hardware interactions, or under new interference patterns. The traditional fault recovery systems are largely based on models that are trained on or pre-ignorance sets that are ineffective in the new realities. As a result, these zero-day faults will normally result in serious performance impairment, service downtime and expensive human interventions. To solve this urgent real-time problem, we provide a deep meta-learning-based solution that focuses specifically on adaptive, fast, and autonomous fault recovery in self-configurable wireless networks called MIND-RECOVER (Meta Intelligent Neural Decoder for REsilient Configurable nEtworks in Real-time). In contrast to traditional learning approaches, which require extensive retraining, MIND-RECOVER introduces a Model-Agnostic Meta-Learning (MAML) paradigm to identify generalised patterns across various fault conditions. This will enable the system to learn quickly in zero-day conditions using limited data points. Combined with deep reinforcement learning, the system will automatically create reconfiguration policies for rerouting, switching channels, or power control based on real-time feedback, and all these processes can be completed in milliseconds. Thorough simulations on synthetically generated and real-world network fault data demonstrate that MIND-RECOVER is more than 89% accurate, with a 91.3% recovery rate and 3.4 times faster recovery initiation than state-of-the-art baseline approaches. In addition, the framework is found to be resilient across different topologies, fault injection rates, and environmental noise, hence it is viable for deployment in real-life next-generation networks. To conclude, the presented meta-learning architecture will be a crucial step toward designing resilient, intelligent, and self-healing wireless access networks that can withstand unpredictable outages and zero-day fault scenarios.

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

Deep Meta-Learning for Zero-Day Fault Recovery in Self-Configurable Wireless Access Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Layth Hussein

摘要

Wireless Access Networks (WANs) and especially self-configurable systems in smart cities, the Army, and mass IoT ecosystems are becoming increasingly popular as a key to the seamless connectivity and automation. However, they are very dynamic and thus prone to zero-day faults - faults or failures never previously experienced, which arise in changing environments, with new hardware interactions, or under new interference patterns. The traditional fault recovery systems are largely based on models that are trained on or pre-ignorance sets that are ineffective in the new realities. As a result, these zero-day faults will normally result in serious performance impairment, service downtime and expensive human interventions. To solve this urgent real-time problem, we provide a deep meta-learning-based solution that focuses specifically on adaptive, fast, and autonomous fault recovery in self-configurable wireless networks called MIND-RECOVER (Meta Intelligent Neural Decoder for REsilient Configurable nEtworks in Real-time). In contrast to traditional learning approaches, which require extensive retraining, MIND-RECOVER introduces a Model-Agnostic Meta-Learning (MAML) paradigm to identify generalised patterns across various fault conditions. This will enable the system to learn quickly in zero-day conditions using limited data points. Combined with deep reinforcement learning, the system will automatically create reconfiguration policies for rerouting, switching channels, or power control based on real-time feedback, and all these processes can be completed in milliseconds. Thorough simulations on synthetically generated and real-world network fault data demonstrate that MIND-RECOVER is more than 89% accurate, with a 91.3% recovery rate and 3.4 times faster recovery initiation than state-of-the-art baseline approaches. In addition, the framework is found to be resilient across different topologies, fault injection rates, and environmental noise, hence it is viable for deployment in real-life next-generation networks. To conclude, the presented meta-learning architecture will be a crucial step toward designing resilient, intelligent, and self-healing wireless access networks that can withstand unpredictable outages and zero-day fault scenarios.