Scholarlink

<p>The rapid growth of Internet of Things (IoT) ecosystems has transformed distributed task scheduling into a large-scale, real-time optimization problem that demands parallel and distributed processing across heterogeneous edge infrastructures. Traditional deep reinforcement learning (DRL) methods offer adaptability but lack formal mechanisms for enforcing hard constraints, while symbolic reasoning ensures policy compliance and interpretability but does not scale to high-concurrency, dynamic environments. This paper proposes NeuroSCALE, a neuro-symbolic optimization framework for scalable and constraint-aware resource management in edge-enabled IoT systems. NeuroSCALE integrates symbolic policy encoding with neural scheduling through three coordinated modules: a symbolic constraint encoder that prunes infeasible actions, a deep policy network trained using proximal policy optimization, and a symbolic projection operator that corrects constraint violations while preserving performance. The architecture supports decentralized execution, heterogeneous edge profiles, and multi-objective trade-offs among latency, energy consumption, and policy adherence, and naturally maps to parallel and distributed execution models. Experiments on a large-scale simulated testbed with 500 IoT devices and 60 edge nodes demonstrate that NeuroSCALE achieves 97.4% constraint satisfaction, reduces average latency to 151.9&#xa0;ms, and maintains task resilience at 98.5% under node failures, while invoking symbolic reconciliation in fewer than 10% of decisions. These results establish NeuroSCALE as a practical, interpretable, and supercomputing-relevant solution for real-time optimization in large-scale edge-IoT systems.</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

Neuroscale: scalable and constraint-aware neuro-symbolic optimization for edge-centric IoT systems

  • 期刊论文

  • 发表日期：

    2026年2月4日

查看全文
  • Waseem Abbass,
  • Nasim Abbas,
  • Uzma Majeed

摘要

The rapid growth of Internet of Things (IoT) ecosystems has transformed distributed task scheduling into a large-scale, real-time optimization problem that demands parallel and distributed processing across heterogeneous edge infrastructures. Traditional deep reinforcement learning (DRL) methods offer adaptability but lack formal mechanisms for enforcing hard constraints, while symbolic reasoning ensures policy compliance and interpretability but does not scale to high-concurrency, dynamic environments. This paper proposes NeuroSCALE, a neuro-symbolic optimization framework for scalable and constraint-aware resource management in edge-enabled IoT systems. NeuroSCALE integrates symbolic policy encoding with neural scheduling through three coordinated modules: a symbolic constraint encoder that prunes infeasible actions, a deep policy network trained using proximal policy optimization, and a symbolic projection operator that corrects constraint violations while preserving performance. The architecture supports decentralized execution, heterogeneous edge profiles, and multi-objective trade-offs among latency, energy consumption, and policy adherence, and naturally maps to parallel and distributed execution models. Experiments on a large-scale simulated testbed with 500 IoT devices and 60 edge nodes demonstrate that NeuroSCALE achieves 97.4% constraint satisfaction, reduces average latency to 151.9 ms, and maintains task resilience at 98.5% under node failures, while invoking symbolic reconciliation in fewer than 10% of decisions. These results establish NeuroSCALE as a practical, interpretable, and supercomputing-relevant solution for real-time optimization in large-scale edge-IoT systems.