An Improved Video Recommendation System for IoT Devices Using Federated Learning

Video recommendation progressively adds new dimensions to the application of Internet of Things (IoT). JointRec, a video recommendation architecture for minority users, introduced federated learning in the JointCloud operation. In our work, the proposed design trains the edge end devices of the local clouds incorporating Federated Swapping (FedSwap) into Federated Averaging (FedAvg) operation during cloud weight aggregation. The implemented Convolutional Neural Network (CNN) model uses Dual Convolutional Probabilistic Matrix Factorization (Dual-CPMF) method in the parallel training of clouds. The communication cost of weighted file transmission is handled with a compression aggregator. After 10 training rounds, this model obtains RMSE = 0.8558 and aggregates 1.43 Gbits validating the prediction accuracy and communication efficiency of the proposed video recommendation system.