A-FedPD: Aligning Dual-Drift is All Federated Primal-Dual Learning Needs

Authors: Yan Sun, Li Shen, Dacheng Tao

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments are conducted on several classical FL setups to validate the effectiveness of our proposed method.
Researcher Affiliation Academia Yan Sun The University of Sydney ysun9899@uni.sydney.edu.au Li Shen Shenzhen Campus of Sun Yat-sen University mathshenli@gmail.com Dacheng Tao Nanyang Technological University dacheng.tao@ntu.edu.sg
Pseudocode Yes Algorithm 1 A-Fed PD Algorithm
Open Source Code Yes We submit our code demo to reproduce the experiments and all hyperparameters can be found in our paper. ... We submit the code demo to reproduce the experiments.
Open Datasets Yes We follow previous work to test the performance of benchmarks on the CIFAR-10 / 100 dataset Krizhevsky et al. [2009].
Dataset Splits No The paper states 'The total dataset of both contain 50,000 training samples and 10,000 test samples of 10 / 100 classes.' for CIFAR-10/100, providing training and test set sizes, but does not explicitly detail a separate validation split or its size/percentage.
Hardware Specification Yes Hardware: NVIDIA Ge Force RTX 2080 Ti
Software Dependencies Yes Platform: Pytorch 2.0.1 Cuda: 11.7
Experiment Setup Yes In each setup, for a fair comparison, we freeze the most of hyperparameters for all methods. We fix total communication rounds T = 800 except for the ablation studies. ... Table 3: Hyperparameters selections of benchmarks.