Federated deep reinforcement learning for network-based predictive edge computing in 6G mobile networks
Аннотация
This paper examines the use of Federated Deep Reinforcement Learning (FDRL) for network load optimization using edge computing in 6G mobile networks. In this work, the model assumes N =10 edge nodes, where each node will use FDRL agents (state space S =5, actions A =3) to determine the offloading/reallocation decision depending on the current and predicted load and resource characteristics. Local training of each FDRL agent is followed by aggregation of the model parameters on the server using the FDRL method, after which the updated global policy is distributed to the nodes. MATLAB simulations ( E =100 episodes) show a reduction in the overall network load/latency from 363 ms to 117 ms (a reduction of 68%). These results demonstrate that FDRL can be used for adaptive offloading in 6G edge networks, given the low volumes of raw data transfer. This article explores the use of Federated Deep Reinforcement Learning (FDRL) to optimize network load using edge computing in 6G mobile networks.
Текст статьи
This paper examines the use of Federated Deep Reinforcement Learning (FDRL) for network load optimization using edge computing in 6G mobile networks. In this work, the model assumes N =10 edge nodes, where each node will use FDRL agents (state space S =5, actions A =3) to determine the offloading/reallocation decision depending on the current and predicted load and resource characteristics. Local training of each FDRL agent is followed by aggregation of the model parameters on the server using the FDRL method, after which the updated global policy is distributed to the nodes. MATLAB simulations ( E =100 episodes) show a reduction in the overall network load/latency from 363 ms to 117 ms (a reduction of 68%). These results demonstrate that FDRL can be used for adaptive offloading in 6G edge networks, given the low volumes of raw data transfer. This article explores the use of Federated Deep Reinforcement Learning (FDRL) to optimize network load using edge computing in 6G mobile networks.