Efficient Federated Learning against Heterogeneous and Non-stationary Client Unavailability

Authors: Ming Xiang, Stratis Ioannidis, Edmund Yeh, Carlee Joe-Wong, Lili Su

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

Reproducibility Variable Result LLM Response
Research Type Experimental We corroborate our analysis with numerical experiments over diversified client unavailability dynamics on real-world data sets.
Researcher Affiliation Academia 1Northeastern University, Boston, MA 2Carnegie Mellon University, Pittsburgh, PA
Pseudocode Yes Algorithm 1: Fed AWE
Open Source Code Yes The code for reproducing our experiments is available at https://github.com/mingxiang12/Fed AWE.
Open Datasets Yes The image classification tasks use CNNs and are based on SVHN [37], CIFAR-10 [26] and CINIC-10 [12] data sets.
Dataset Splits No The paper mentions 'train images' and 'test images' with their counts for SVHN, CIFAR-10, and CINIC-10 datasets. It also states that learning rates are 'searched, based on the best performance after 500 global rounds', implying a validation process, but does not explicitly define a separate 'validation' split or its size.
Hardware Specification Yes The simulations are performed on a private cluster with 64 CPUs, 500 GB RAM and 8 NVIDIA A5000 GPU cards.
Software Dependencies Yes We code the experiments based on Py Torch 1.13.1 [40] and Python 3.7.16.
Experiment Setup Yes Table 6 specifies details of the structures of the convolutional neural network and training. ... The initial local learning rate η0 and the global learning rate ηg are searched, based on the best performance after 500 global rounds, over two grids {0.1, 0.05, 0.01, 0.005, 0.001, 0.0005} and {0.5, 1, 1.5, 5, 10, 50}, respectively.