FedDUAL: A Dual-Strategy with Adaptive Loss and Dynamic Aggregation for Mitigating Data Heterogeneity in Federated Learning

This repository contains the code for the paper "FedDUAL: A Dual-Strategy with Adaptive Loss and Dynamic Aggregation for Mitigating Data Heterogeneity in Federated Learning".

Dependencies

• Tensorflow = 2.10.0
• scikit-learn = 1.3.2

Data Preparing

To divide the dataset into the required no. of clients, run Data Prepration.py and choose the required dataset (CIFAR10, CIFAR100 or FMNIST) and then change the degree of heterogenity (alpha) as required. you will get the desired distribution for each client.

Model Structure

To choose the appropriate model, run Models.py and choose the required model for each of the dataset.

Run FedDUAL

After done with above process, you can run the FedDUAL, our proposed method by running FedDUAL.py.

Evaluation

After federated training, run Evaluation.py to acess the evaluation metrics such as accuracy, precision, recall etc.