WEIGHTED AVERAGE FEDERATED LEARNING BASED ON NEURAL NETWORK TRAINING LOSS: abstract simplified (17672533)

The abstract describes a method of wireless communication that involves a user equipment (UE) computing updates to an artificial neural network during a federated learning process. These updates can be gradients or updated model parameters. The method also involves recording the training loss observed while training the neural network at that epoch of the federated learning process. Finally, the method includes transmitting these updates to a federated learning server, which is responsible for aggregating the gradients based on the training loss.