FedExP: Speeding Up Federated Averaging via Extrapolation

Authors: Divyansh Jhunjhunwala, Shiqiang Wang, Gauri Joshi

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that Fed Ex P consistently converges faster than Fed Avg and competing baselines on a range of realistic FL datasets.
Researcher Affiliation Collaboration 1Carnegie Mellon University, 2IBM Research
Pseudocode Yes Algorithm 1 Proposed Algorithm: Fed Ex P
Open Source Code Yes Our code is available at the following link https://github.com/Divyansh03/Fed Ex P.
Open Datasets Yes For realistic FL tasks, we consider image classification on the following datasets i) EMNIST (Cohen et al., 2017), ii) CIFAR-10 (Krizhevsky et al., 2009), iii) CIFAR-100 (Krizhevsky et al., 2009), iv) CINIC-10 (Darlow et al., 2018).
Dataset Splits No For EMNIST we use the federated version of EMNIST available at Caldas et al. (2019), which is naturally partitioned into 3400 clients. The number of training and test samples is 671,585 and 77,483 respectively. For CIFAR-10/100...the number of training examples and test examples is 50,000 and 10,000 respectively. The paper clearly defines training and test splits, but not a separate validation split for reproduction.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, memory, or cloud computing instances used for running experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers, such as 'Python 3.8' or 'PyTorch 1.9'.
Experiment Setup Yes For our baselines, we find the best performing ηg and ηl by grid-search tuning. For Fed Ex P we optimize for ϵ and ηl by grid search. We fix the number of participating clients to 20, minibatch size to 50 and number of local updates to 20 for all experiments. In Appendix D, we provide additional details and results, including the best performing hyperparameters...