Wirelessly Powered Federated Edge Learning: Optimal Tradeoffs Between Convergence and Power Transfer

Wireless for Machine Learning: A Survey

“…Summary [32] Considers update staleness and update drift to develop a joint client selection and resource block allocation scheme. Energy Harvesting/Power Transfer [185] Joint batch size selection, clock frequency optimization, and learning-wireless power transfer tradeoff. [148] Joint local number of iterations optimization and time slot allocation to transmit, compute and harvest energy.…”

“…However, how to allow the devices to harvest sufficient energy to train a FL model while not substantially increasing the communication round time is largely an open question. The use of energy harvesting for FL is highly novel and to the best of our knowledge, there are only three works in the literature [62], [148], [185].…”

“…In [185], the authors consider an FL application in which a wireless network uses power-beacons to transfer radio frequency energy. The key components of the work are the distributed gradient estimation, local-computation optimization, and optimal learning-wireless power transfer tradeoff.…”

“…With the increasing use of IoT devices for monitoring applications, the importance of energy harvesting for FL is quickly increasing. The works we discussed [62], [148], [185] are the first attempts to analyze this emerging field, but substantial work is still necessary. Specifically, the impact of its application with bandwidth limited transmissions, such as narrowband IoT which limits the transmission rate for the IoT devices.…”

Reconfigurable Intelligent Surface-Enabled Federated Learning for Power-Constrained Devices

“…In particular, in [11], a hybrid radio frequency (RF)/visible light communication (VLC) scenario was considered, in which the VLC link over the downlink is leveraged for energy harvesting purposes, while the RF is used for the local model updates transmission. From a similar perspective, the authors in [12] applied the time-switching paradigm of WPT to an FL system, in which they studied the trade-off between learning and WPT, and further optimized the MU clock frequency for improved utilization of harvested energy for model evaluation. From a similar point of view, the authors in [14] minimized the mean squared error (MSE) by optimizing the aggregation beamforming and consumed energy in a vehicular energy-limited network.…”