Distributed learning on edges aims at training the AI model collaboratively in a network of edge devices via frequent model aggregations. Achieving the desired training performance requires the aggregation structure and frequency to fit well with the dynamic edge environment. Existing works often consider the optimization of either aggregation structure or frequency, assuming that the edge environment is stable and deterministic. In this paper, we propose a novel approach,AutoSF, to automatically optimize the aggregation structure and frequency jointly in dynamic edge computing so as to minimize the global loss function. The main idea of AutoSF is that when the edge environment changes, the automated machine learning approach is triggered to find out the near-optimal aggregation structure and frequency that adapt to time-varying edge resources. When the environment keeps unchanged, a heuristic approach is used to tune the aggregation structure and frequency to further tame the heterogeneity caused by data distributions. We validate the effectiveness of AutoSF via numerical experiments with real datasets on our self-developed edge computing testbed. Evaluation results demonstrate that AutoSF outperforms the benchmark approaches by up to 16.3× speedups in convergence speed and 31.0$\%$increases in training accuracy.

Federated Learning (FL) has been widely used for distributed machine learning in edge computing. In FL, the model parameters are iteratively aggregated from the clients to a central server, which is inclined to be the communication bottleneck and single point of failure. To solve these drawbacks, hierarchical model training frameworks like Hierarchical Federated Learning (HFL) and E-Tree learning have been proposed. One of the most challenging problems in the hierarchical model training framework is optimizing the aggregation frequencies of the edge devices at various levels. Because, in an edge computing environment, heterogeneity in the resource can introduce synchronization delays caused by waiting for slow workers and significantly impact the training performance. This paper tackles the problem with weak synchronization where edge devices on the same level have different frequencies on local updates and/or model aggregations. Existing works based on weak synchronization lack solutions to quantitatively determine the aggregation frequencies of each edge device. Thus, we propose a resource-based aggregation frequency controlling method, termed RAF, which determines the optimal aggregation frequencies of edge devices to minimize the loss function according to heterogeneous resources. Our proposed method can alleviate the waiting time and fully utilize the resources of the edge devices. Besides, RAF dynamically adjusts the aggregation frequencies at different phases during the model training to achieve fast convergence speed and high accuracy. We evaluated the performance of RAF via extensive experiments with real datasets on our self-developed edge computing testbed. Evaluation results demonstrate that RAF outperforms the benchmark approaches in terms of learning accuracy and convergence speed.