Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
PA3Fed: Period-Aware Adaptive Aggregation for Improved Federated Learning
Authors: Chengxiang Huang, Bingyan Liu
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that our method improves accuracy by up to 15% and significantly enhances stability. We conduct extensive experiments on various datasets and models and validate the excellent performance of our method by incorporating it into several state-of-the-art approaches, including Fed Avg (Mc Mahan et al. 2017), Fed Prox (Li et al. 2020), Fed Nova (Wang et al. 2020b), VRL-SGD (Liang et al. 2019), and Fed Mut (Hu et al. 2024). The results indicate that our method can improve performance by up to 15% under the high non IID conditions compared to the original methods. In addition, we conduct a series of in-depth empirical analyses to prove the effectiveness of the proposed method. |
| Researcher Affiliation | Academia | Chengxiang Huang1, Bingyan Liu1 * 1Beijing University of Posts and Telecommunications EMAIL |
| Pseudocode | Yes | As depicted in Figure 2 and detailed in Algorithm 1, in this paper, we propose the PA3Fed approach, which dynamically adjusts the aggregation coefficient based on the performance variations of local clients and the stage of training. ... Algorithm 1: Pipeline of PA3Fed |
| Open Source Code | No | The paper does not explicitly provide a link to source code, nor does it state that the code is publicly available or included in supplementary materials. |
| Open Datasets | Yes | We employ three of the most commonly used image classification datasets in federated learning: CIFAR-10 (Krizhevsky, Hinton et al. 2009), CIFAR-100 (Krizhevsky, Hinton et al. 2009), and Fashion-MNIST (Xiao, Rasul, and Vollgraf 2017). |
| Dataset Splits | No | The paper describes how data is distributed among K clients using a Dirichlet distribution parameterized by h to simulate non-i.i.d. conditions. It mentions 'Final Test Accuracy' implying a test set, but it does not provide specific percentages or counts for training, validation, and test splits for the overall datasets or for individual client datasets beyond the non-i.i.d partitioning strategy. It does not explicitly state the use of standard train/test splits for the named datasets. |
| Hardware Specification | Yes | All our experiments are simulated and executed on a server configured with three NVIDIA Ge Force RTX 3090 GPUs, 48 Intel Xeon CPU cores, and 128GB of RAM. |
| Software Dependencies | No | The paper does not provide specific version numbers for any software libraries, frameworks, or programming languages used in the experiments. |
| Experiment Setup | Yes | The batch size is set to 32, with an initial learning rate of 0.01 that decays by a constant factor of 0.8 after each communication round. Additionally, we applied a momentum of 0.9 and a weight decay of 10 5. ... Each client conducts 5 local training rounds and the global model is trained for a total of 250 rounds. The detection threshold of CLP is set to 0.01. The hyperparameter β, which controls the modification degree of the aggregation, is fixed at 0.3. |