EBS-CFL: Efficient and Byzantine-robust Secure Clustered Federated Learning
Authors: Zhiqiang Li, Haiyong Bao, Menghong Guan, Hao Pan, Cheng Huang, Hong-Ning Dai
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct theoretical analysis and extensive experiments to evaluate our proposed EBS-CFL. Through theoretical analysis, we explain the communication and computational complexity of our scheme. Moreover, we conducted experiments involving multiple variables related to communication and computation overhead, conducting detailed data analysis of the entire process of aggregation stage. And we incorporate experiments involving typical Byzantine attacks in FL to validate the robustness of our proposed EBS-CFL. Finally, through extensive experiments, we demonstrate the efficiency, effectiveness, and robustness of our approach. We carry out experiments on MNIST (Deng 2012), CIFAR10, and CIFAR100 (Krizhevsky and Hinton 2009). |
| Researcher Affiliation | Academia | 1 East China Normal University 2 Fudan University 3 Hong Kong Baptist University EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1: Clustered Model Update({θi}m i=1, D, b, η, T) Input: initialization {θi}m j=1; training datasets D; batch size b; learning rate η; number of iterations T. 1: {θ0 i }m i=1 {θi}m i=1 2: Loss(D; θ0 j) min({Loss(Db; θ0 i }m i=1)) 3: for t = 1 to T do 4: Randomly sample a batch Db from D. 5: θt j θt 1 j η Loss(Db; θt 1 j ) 6: end for 7: return j, θT j θ0 j |
| Open Source Code | Yes | Code https://github.com/Lee-VA/EBS-CFL |
| Open Datasets | Yes | We carry out experiments on MNIST (Deng 2012), CIFAR10, and CIFAR100 (Krizhevsky and Hinton 2009). |
| Dataset Splits | No | The paper mentions simulating "data heterogeneity under a non-i.i.d. setting, we use Dirichlet s distribution (Hsu, Qi, and Brown 2019), where α controls data dispersion" and evaluating "at a high adversarial rate of 40%". This describes the data generation and attack parameters, but does not provide specific train/validation/test splits (e.g., 80/10/10 percentages or sample counts) needed to reproduce data partitioning. |
| Hardware Specification | No | The paper discusses communication and computational overhead and measures "consumed time" for efficiency but does not provide specific details about the CPU, GPU models, or other hardware used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., programming languages, libraries, or frameworks with their respective versions) used for the implementation or experiments. |
| Experiment Setup | No | The paper defines abstract inputs for Algorithm 1 such as "batch size b", "learning rate η", and "number of iterations T", but it does not provide concrete numerical values for these hyperparameters or other specific training configurations used in the experimental setup section. |