Scholarlink

Peer-to-peer (P2P) computation offloading has an Internet of Things (IoT) device to nearby device with accessible resources. This process involves a device-to-device (D2D) communication and minimize dependency on centralized cloud servers and enhance the effectiveness of task execution. P2P offloading increase resource utilization by sharing the workload over numerous devices which minimize energy consumption. However, the inefficient allocation of computational tasks over IoT device cause high delay and energy because of lack of coordination and task distribution. In this research, the multi-agent deep reinforcement learning with stochastic gradient descent (MADRL-SGD) is proposed for P2P Computation Offloading in IoT Edge Computing. It makes effective and dynamic tasks allocation by applying multi-agent collaboration which helps to minimize energy and latency. The SGD provides rapid convergence and enhanced learning effectiveness in IoT edge computing. MADRL adapts to heterogenous device by varying workloads and capabilities which increase overall system performance. The proposed MADRL-SGD achieves a low average total cost of 0.5 for task arrival probability compared to existing methods like deep Q-network (DQN) and DRL, respectively.

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

Multi-agent Deep Reinforcement Learning with Stochastic Gradient Descent for Peer-to-Peer Computation Offloading in IoT Edge Computing

  • 章节

  • 发表日期：

    2026年

查看全文
  • K. V. Sheelavathy,
  • V. Prabhudeva,
  • D. Kannan,
  • Debashis Rudra Sarma

摘要

Peer-to-peer (P2P) computation offloading has an Internet of Things (IoT) device to nearby device with accessible resources. This process involves a device-to-device (D2D) communication and minimize dependency on centralized cloud servers and enhance the effectiveness of task execution. P2P offloading increase resource utilization by sharing the workload over numerous devices which minimize energy consumption. However, the inefficient allocation of computational tasks over IoT device cause high delay and energy because of lack of coordination and task distribution. In this research, the multi-agent deep reinforcement learning with stochastic gradient descent (MADRL-SGD) is proposed for P2P Computation Offloading in IoT Edge Computing. It makes effective and dynamic tasks allocation by applying multi-agent collaboration which helps to minimize energy and latency. The SGD provides rapid convergence and enhanced learning effectiveness in IoT edge computing. MADRL adapts to heterogenous device by varying workloads and capabilities which increase overall system performance. The proposed MADRL-SGD achieves a low average total cost of 0.5 for task arrival probability compared to existing methods like deep Q-network (DQN) and DRL, respectively.