A Federated Incremental Learning Algorithm Based on Dual Attention Mechanism

Federated incremental learning best suits the changing needs of common Federal Learning (FL) tasks. In this area, large sample client dramatically influences final model training results, and unbalanced features are challenging to capture. paper, a federated framework is designed; firstly, part data preprocessed obtain initial global model. Secondly, help get importance whole each client, enhance performance capture critical information feature, channel attention neural network designed on side, aggregation algorithm based feature mechanism server side. Experiments standard datasets CIFAR10 CIFAR100 show that proposed accuracy has good premise realizing learning.