AI 在 6G 系統和元宇宙的邊緣計算中至關重要，透過在本地設備、邊緣伺服器和雲端之間實現即時適應和高效任務分配來最佳化資源管理並提升用戶體驗。它確保了無縫連接，降低了延遲，並提高了整體系統性能和擴展性。O-RAN 架構通過開放接口和 AI/ML 驅動的自動化來增強行動網路的靈活性和效率，用於資源管理和切換預測等任務，確保了擴展性和性能。挑戰包括選擇最佳的 AI/ML 模型和部署位置，這對於維持低延遲和有效的資源分配至關重要。在 5G-MEC 中，有效的資源分配著重於最小化延遲和最大化數據傳輸速率，以確保高 QoS，從而改善用戶體驗和系統性能。先進技術如深度強化學習（DRL）被提出來處理 6G 支援的移動邊緣計算（MEC）和元宇宙應用中的動態和資源密集型需求，最佳化資源分配並減少能耗和計算延遲。我們將 meta-learning 整合到 DRL 算法中，以更好地分配 5G MEC 系統中的資源，增強其適應性和有效性。通過為 O-RAN 添加 attention 機制和層次結構，我們顯著改進了反饋、靈活性和資源管理，確保了多樣化應用的卓越 QoS。與 meta-learning 方法相比，meta-attention 方法持續展示出優越的性能，平均滿意度提高了 22%。;AI is vital in edge computing for 6G systems and the metaverse, optimizing resource management and enhancing user experiences by enabling real-time adaptation and efficient task distribution across local devices, edge servers, and the cloud. It ensures seamless connectivity, reduces latency, and improves overall system performance and scalability.The O-RAN architecture enhances mobile network flexibility and efficiency through open interfaces and AI/ML-driven automation for tasks like resource management and handover prediction, ensuring scalability and performance. Challenges include selecting optimal AI/ML models and deployment locations, crucial for maintaining low latency and effective resource allocation.Effective resource allocation in 5G-MEC focuses on minimizing latency and maximizing data rates to ensure high QoS, thereby improving user experience and system performance. Advanced techniques like deep reinforcement learning (DRL) are proposed to handle the dynamic and resource-intensive demands of 6G-enabled mobile edge computing (MEC) and Metaverse applications, optimizing resource allocation and reducing energy use and computation delays.We integrated Meta-Learning into DRL algorithms for better resource allocation in 5G MEC systems, enhancing adaptability and effectiveness. By adding an attention mechanism and a hierarchical structure for O-RAN, we significantly improved feedback, flexibility, and resource management, ensuring superior QoS for diverse applications.The Meta-Attention method consistently demonstrates superior performance, with an average satisfaction increase of 22\% compared to Meta-Learning approaches.
