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 and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Sharp Bounds for Sequential Federated Learning on Heterogeneous Data

Authors: Yipeng Li, Xinchen Lyu

JMLR 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results validate the counterintuitive theoretical finding.
Researcher Affiliation Academia Yipeng Li EMAIL National Engineering Research Center for Mobile Network Technologies Beijing University of Posts and Telecommunications Beijing, 100876, China; Xinchen Lyu EMAIL National Engineering Research Center for Mobile Network Technologies Beijing University of Posts and Telecommunications Beijing, 100876, China
Pseudocode Yes Algorithm 1: Sequential FL; Algorithm 2: Parallel FL
Open Source Code Yes The code is available at https://github.com/liyipeng00/SFL.
Open Datasets Yes We use two data sets a9a and w8a from LIBSVM library (Chang and Lin, 2011). We consider the common CV tasks, with data sets including Fashion-MNIST (Xiao et al., 2017), CIFAR-10 (Krizhevsky et al., 2009), CINIC-10 (Darlow et al., 2018).
Dataset Splits Yes We partition them into M = 1000 clients by Extended Dirichlet strategy (Li and Lyu, 2023), with each client containing data samples from C = 1, 2 labels. ... We partition the training sets of Fashion-MNIST/CIFAR-10/CINIC-10 into 500/500/1000 clients by Extended Dirichlet strategy (Li and Lyu, 2023), with each client containing data samples from C = 1, 2, 5 labels. We spare the original test sets for computing test accuracy.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. It mentions using 'a CNN model' and 'a VGG-9 model' but no hardware specifics.
Software Dependencies No The paper mentions 'The local solver is SGD' and 'We apply gradient clipping to both algorithms' but does not specify any software libraries or frameworks with version numbers (e.g., Python, PyTorch, TensorFlow, CUDA versions).
Experiment Setup Yes We set the number of local steps to K = 5, the number of participating clients to S = 10, and the mini-batch size to 8. The local solver is SGD with learning rate being constant, momentum being 0 and weight decay being 0. We tune the learning rate by the grid search. ... We fix the number of participating clients per round to S = 10. We fix the number of local update steps to K = 5 and the mini-batch size to 20 ... We apply gradient clipping to both algorithms and tune the learning rate by grid search with a grid of {10 2.5, 10 2.0, 10 1.5, 10 1.0, 10 0.5}.