Data analysis algorithm for internet of things based on federated learning with optical technology

As the Internet of Things (IoT) progresses, federated learning (FL), a decentralized machine learning framework that preserves every participant's data privacy, has grown in prominence. However the IoT data possessed by corporations and enterprises frequently has different distributed properties (Non-IID), which has a negative influence on their federated learning. Throughout the local training stage, this issue makes client forget about global information, which therefore slows convergence in general and reduce accuracy. The suggested technique called FedARD, which depends on relationship-based insight distillation, to improve the mining of higher grade global knowledge through local algorithms from a superior dimensions viewpoint over their term of local training in order to maintain global knowledge and prevent forgetting. In order for students to further efficiently acquire global knowledge, it also established an entropy-wise adaptive weights module to control the proportional of loss in single sample knowledge distillation against relational knowledge distillation. FedARD performed stronger than other sophisticated FL approaches in terms of convergence speed along with classification accuracy, as determined by a set of studies on CIFAR10 as well as CIFAR100.