关键词：
Cooperative learning
Neurons
Algorithms
Privacy
Servers
Keyboards
Breast cancer
Neural networks
Artificial intelligence
Computer science
Oncology
摘要：
Federated learning is a distributed machine learning method that enables model training on client devices, which protects the privacy of user data. Infederated learning, instead of passing clients’ data to a server, clients generated models are transferred to the central server, which in turn aggregates all models into a global model. The global model then goes back to clients and continues the same process to improve model performance. We proposed three central server model aggregation techniques namely, all model averaging(AMA), one model selection (OMS), and best models averaging (BMA) in this thesis. We adopted neural network models in this study and implemented the aggregation by averaging the neural network model coefficients. Multilayer-perceptron and convolutional neural network models are tested with four unstructured data sets, handwritten-digits, fingerprints, breast cancer data set, and CIFAR-10 data set. The proposed aggregation techniques achieved an average of 99.46% model accuracy for handwritten-digits, 82.98% model accuracy for fingerprints, 66.66% mode accuracy for breast cancer, and 60.91% model accuracy for CIFAR-10 datasets. In addition, the CIFAR-10 data set is used in a federated learning framework to validate the chosen models and aggregation techniques. The AMA technique works better than the other two techniques for every data set in our study. The OMS technique outperforms other methods for every data set in some cases. However, BMA did not perform well for the fingerprint data set but outperform other methods in some cases for both handwritten-digits and breast cancer datasets. We have also worked on clients’ dropout scenario in federated learning and two cluster-based federated learning frameworks using the AMA aggregation technique. The experiment results suggest that the proposed model aggregation techniques work well in the federated learning frameworks.
欢迎来到三峡大学图书馆！
