FedBR: Improving Federated Learning on Heterogeneous Data via Local Learning Bias Reduction
Authors: Yongxin Guo, Xiaoying Tang, Tao Lin
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted several experiments to test Fed BR and found that it consistently outperforms other SOTA FL methods. Both of its components also individually show performance gains. |
| Researcher Affiliation | Academia | 1School of Science and Engineering, The Chinese University of Hong Kong, Shenzhen, Guangdong 518172, China. 2The Shenzhen Institute of Artificial Intelligence and Robotics for Society 3The Guangdong Provincial Key Laboratory of Future Networks of Intelligence, The Chinese University of Hong Kong (Shenzhen), Shenzhen 518172, China 4Research Center for Industries of the Future, Westlake University 5School of Engineering, Westlake University. Correspondence to: Xiaoying Tang <tangxiaoying@cuhk.edu.cn>. |
| Pseudocode | Yes | The proposed Fed BR is summarized in Algorithm 1. ... Algorithm 1 Algorithm Framework of Fed BR ... Algorithm 2 Construct Augmentation Data ... Algorithm 3 Construct Augmentation Data by Proxy Data |
| Open Source Code | Yes | Our code is available at https: //github.com/lins-lab/fedbr. |
| Open Datasets | Yes | We examine all algorithms on Rotated MNIST, CIFAR10, and CIFAR100 datasets. ... We split the datasets following the idea introduced in (Yurochkin et al., 2019; Hsu et al., 2019; Reddi et al., 2021), where we leverage the Latent Dirichlet Allocation (LDA) to control the distribution drift with parameter α. |
| Dataset Splits | Yes | We split the datasets following the idea introduced in (Yurochkin et al., 2019; Hsu et al., 2019; Reddi et al., 2021), where we leverage the Latent Dirichlet Allocation (LDA) to control the distribution drift with parameter α. ... Unless specially mentioned, we split Rotated MNIST, CIFAR10, and CIFAR100 to 10 clients and set α = 0.1. |
| Hardware Specification | No | The paper mentions the use of models like VGG11 and CCT but does not specify any hardware details such as GPU models, CPU types, or other computing resources used for the experiments. |
| Software Dependencies | No | The paper specifies hyperparameters and optimizers used (e.g., SGD, learning rates, momentum) but does not provide specific version numbers for software libraries, programming languages (e.g., Python, PyTorch), or other ancillary software dependencies. |
| Experiment Setup | Yes | We use SGD optimizer and set the learning rate to 0.001 for Rotated MNIST, and 0.01 for other datasets. The local batch size is set to 64 for Rotated MNIST, and 32 for other datasets. ... Each communication round includes 50 local iterations, with 1000 communication rounds for Rotated MNIST and CIFAR10, 800 communication rounds for CIFAR100, and 400 communication rounds for PACS. ... We utilize a four-layer CNN for MNIST, VGG11 for CIFAR10 and PACS, and CCT ... for CIFAR100. ... By default, we set τ1 = τ2 = 2.0, the weight of contrastive loss µ = 0.5, and the weight of Aug Mean λ = 1.0 on MNIST and CIFAR100, λ = 0.1 on CIFAR10 and PACS. |