Time-Sensitive Learning for Heterogeneous Federated Edge Intelligence

Real-time machine learning (ML) has recently attracted significant interest due to its potential to support instantaneous learning, adaptation, and decision making in a wide range of application domains. In this paper, we investigate real-time ML in a federated edge intelligence (FEI) system. We propose a time-sensitive federated learning (TS-FL) framework to minimize the overall run-time for training a shared ML model with desirable accuracy. Training acceleration solutions for both TS-FL with synchronous coordination (TS-FL-SC) and asynchronous coordination (TS-FL-ASC) are developed. To address the straggler effect in TS-FL-SC, we develop an analytical solution to characterize the impact of selecting different subsets of edge servers on the model training time. A server dropping-based solution is proposed to allow slow-performance edge servers to be removed from participating in the model training if their impacts on the model accuracy are limited. A joint optimization algorithm is proposed to minimize the overall model training time by selecting participating edge servers, local epoch number, and data batch sizes. We also develop an analytical expression to characterize the impact of staleness effect of asynchronous coordination on the training time of TS-FL-ASC. We propose a load forwarding-based solution that allows slow edge servers to offload part of their training samples to trusted edge servers with fast processing capability. We develop a hardware prototype to evaluate the performance of our proposed solutions. Experimental results show that TS-FL-SC and TS-FL-ASC can provide up to 63% and 28% of reduction in the overall model training time, respectively, compared with traditional solutions.