Distributed Graph Neural Network (DGNN) is a powerful tool in large-scale graph representation learning. However, high data-transfer overhead among workers in a DGNN training job confines its scalability and thus the overall performance. Vertex-wise historical embedding methods have demonstrated high potential to alleviate the problems, but still suffer from severe accuracy loss and limited performance scalability, which has been attributed to the information loss of critical channels in historical vertices and redundant information in local channels. This article explores the optimization of channel level and construct a quantitative accuracy model for channel-wise historical embedding. We propose Sift, a novel DGNN training framework, supporting channel-wise partial historical embedding with accuracy guarantee. Sift has three components: a historical embedding evaluator with channel-wise quantitative accuracy model, a sawtooth-like matrix rearrangement for accelerating message passing, and a hybrid parallel framework for overlapping communication overhead. Comprehensive experimental results show that Sift achieves near-linear parallel convergence speedup, outperforming the state-of-the-art baselines by up to 72% in total training performance and up to 21% in convergence speed.

This paper aims to propose a new event-triggered distributed model predictive control framework based on an adaptive dual-horizon (prediction and control horizon) mechanism for multi-agent cyber-physical systems (CPS) with additive disturbances. Firstly, the control horizon is introduced into the constrained optimal control problem (OCP) of multi-agent CPS to reduce the number of independent variables and improve the rapidity of the algorithm. Secondly, a periodic event-triggered mechanism with Zeno-free behavior is designed based on state error information, which can more efficiently reduce the frequency of solving OCP. Then, adaptive prediction horizon and control horizon contraction mechanisms are designed, effectively reducing the computational complexity of solving OCP for the controller at the triggering moment. In addition, sufficient conditions are provided through theoretical analysis to guarantee the recursive feasibility of algorithm and the stability of the closed-loop system. Finally, the effectiveness of the proposed algorithm is verified through simulation and experimentation example of networked multi-robot system, and the results showed that the algorithm can effectively reduce communication resource consumption and computational complexity in solving OCP while ensuring the expected cooperative control effect.

This paper designs a dynamic event-triggered model predictive control (EMPC) for constrained non-holonomic robots with disturbances, with the aim of reducing the frequency and computational complexity of solving optimal control problems (OCPs) to alleviate computational burden while ensuring system stability. Firstly, the prediction horizon and control horizon are considered separately in the OCP, and an EMPC algorithm based on the adaptive prediction horizon and control horizon mechanism is designed. Then, sufficient conditions are provided for guaranteeing the recursive feasibility of the proposed algorithm, closed-loop stability, and the Zeno-free behavior. Finally, the effectiveness of the proposed method is verified through simulation and comparison.