FedMDFG: Federated Learning with Multi-Gradient Descent and Fair Guidance

Authors: Zibin Pan, Shuyi Wang, Chi Li, Haijin Wang, Xiaoying Tang, Junhua Zhao

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, extensive experiments in several FL scenarios verify that Fed MDFG is robust and outperforms the SOTA FL algorithms in convergence and fairness. We evaluate the performance of Fed MDFG on four public datasets: CIFAR-10, CIFAR-100 (Krizhevsky and Hinton 2009), MNIST(Le Cun et al. 1998), and Fashion MNIST (FMNIST) (Xiao, Rasul, and Vollgraf 2017)
Researcher Affiliation Academia 1 The School of Science and Engineering, The Chinese University of Hong Kong, Shenzhen, China 2 The Shenzhen Institute of Artificial Intelligence and Robotics for Society 3 The Guangdong Provincial Key Laboratory of Future Networks of Intelligence
Pseudocode Yes Algorithm 1: Fed MDFG
Open Source Code Yes The source code is available at https://github.com/zibinpan/Fed MDFG.
Open Datasets Yes We evaluate the performance of Fed MDFG on four public datasets: CIFAR-10, CIFAR-100 (Krizhevsky and Hinton 2009), MNIST(Le Cun et al. 1998), and Fashion MNIST (FMNIST) (Xiao, Rasul, and Vollgraf 2017), where the training/testing data are already split.
Dataset Splits No We evaluate the performance of Fed MDFG on four public datasets: CIFAR-10, CIFAR-100 (Krizhevsky and Hinton 2009), MNIST(Le Cun et. al. 1998), and Fashion MNIST (FMNIST) (Xiao, Rasul, and Vollgraf 2017), where the training/testing data are already split.
Hardware Specification Yes All experiments are implemented on a server with Intel(R) Xeon(R) Silver 4216 CPUs and 2 NVidia(R) 3090 GPUs.
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries used in the experiments.
Experiment Setup Yes Table 1 list the hyper-parameters of various techniques we verify. The first one of each parameter set is regarded as the default for each algorithm. All clients utilize Stochastic Gradient Descent (SGD) on local datasets with the learning rate η {0.01, 0.05, 0.1} and decay of 0.999 per round, and take the best performance of each method in comparison. We average the results in 5 runs with different random seeds.