Digital twins (DTs) will revolutionize network autonomy. Recent studies have promoted the idea of a DT-native 6G network, deeply integrating DTs into mobile network architectures to improve the timeliness of physical–digital synchronization and network optimizations. However, DTs have mainly acted just as a tool for network autonomy, leading to a gap regarding the ultimate goal of network self-evolution. This paper analyzes future directions concerning DT-native networks. Specifically, the proposed architecture introduces a key concept called “future shots” that gives accurate network predictions under different time scales of self-evolution strategies for various network elements. To realize the future shots, we propose a long-term hierarchical convolutional graph attention model for cost-effective network predictions, a conditional hierarchical graph neural network for strategy generation, and methods for efficient small-to-large-scale interactions. The architecture is expected to facilitate high-level network autonomy for 6G networks.
While digital twin networks (DTNs) can potentially estimate network strategy performance in pre-validation environments, they are still in their infancy for split learning (SL) tasks, facing challenges like unknown non-i.i.d. data distributions, inaccurate channel states, and misreported resource availability across devices. To address these challenges, this paper proposes a TransNeural algorithm for DTN pre-validation environment to estimate SL latency and convergence. First, the TransNeural algorithm integrates transformers to efficiently model data similarities between different devices, considering different data distributions and device participate sequence greatly influence SL training convergence. Second, it leverages neural network to automatically establish the complex relationships between SL latency and convergence with data distributions, wireless and computing resources, dataset sizes, and training iterations. Deviations in user reports are also accounted for in the estimation process. Simulations show that the TransNeural algorithm improves latency estimation accuracy by and convergence estimation accuracy by compared to traditional equation-based methods.
Digital twin (DT) will revolute network autonomy. Current studies have promoted DT-native 6G network by deeply integrate DT into mobile network architectures to improve the timeless of physical-digital synchronization and network optimizations. However, DT has mainly acted as just a tool for network autonomy, leading a gap towards the ultimate goal of network self-evolution. This paper analyzes the future direction of the DT-native network. Specifically, the proposed architecture introduces a key concept called "future shots", which gives accurate network predictions under different time scales of self-evolution strategies for various network elements. To realize the future shots, we propose a long-term hierarchical convolutional graph attention model for cost-effectively network predictions, a conditional hierarchical graph neural network for strategy generations, and methods for efficient small-large scale interactions. The architecture is expected to facilitate high-level network autonomy for 6G networks.
As split learning (SL) becomes popular for addressing data privacy and resource limitation issues in edge model training process, it is crucial to pre-validate whether the extensive use of wireless and computing resources in SL can achieve the expected training latency and convergence. While digital twin network (DTN) can potentially estimate network strategy performance in pre-validation environments, they are still in their infancy for SL tasks, facing challenges like unknown non-i.i.d. data distributions, inaccurate channel states, and misreport resource availability across devices. To address these challenges, this paper proposes a TransNeural algorithm for DTN pre-validation environment to estimate SL latency and convergence. First, the TransNeural algorithm integrates transformers to efficiently model data similarities between different devices, considering different data distributions and device participate sequence greatly influence SL training convergence. Second, it leverages neural network to automatically establish the complex relationships between SL latency and convergence with data distributions, wireless and computing resources, dataset sizes, and training iterations. Deviations in user reports are also accounted for in the estimation process. Simulations show that the TransNeural algorithm improves latency estimation accuracy by 13.44% and convergence estimation accuracy by 116.74% compared to traditional equation-based methods.