Asynchronous Semi-Decentralized Federated Edge Learning for Heterogeneous Clients
This work addresses efficiency issues in federated edge learning for mobile edge networks, but it appears incremental as it builds on existing architectures with a focus on handling heterogeneity.
The paper tackles the problem of device heterogeneity in federated edge learning by proposing an asynchronous semi-decentralized algorithm with a staleness-aware aggregation scheme, resulting in faster convergence and better learning performance as demonstrated in simulations.
Federated edge learning (FEEL) has drawn much attention as a privacy-preserving distributed learning framework for mobile edge networks. In this work, we investigate a novel semi-decentralized FEEL (SD-FEEL) architecture where multiple edge servers collaborate to incorporate more data from edge devices in training. Despite the low training latency enabled by fast edge aggregation, the device heterogeneity in computational resources deteriorates the efficiency. This paper proposes an asynchronous training algorithm for SD-FEEL to overcome this issue, where edge servers can independently set deadlines for the associated client nodes and trigger the model aggregation. To deal with different levels of staleness, we design a staleness-aware aggregation scheme and analyze its convergence performance. Simulation results demonstrate the effectiveness of our proposed algorithm in achieving faster convergence and better learning performance.