Problem-Parameter-Free Federated Learning

Authors: Wenjing Yan, Kai Zhang, Xiaolu Wang, Xuanyu Cao

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive empirical evidence validates the efficacy of our approach. We conduct empirical evaluations to validate our theoretical findings and the efficacy of our algorithms. In this section, we present experiments on two real-world datasets: EMNIST (Cohen et al., 2017) and CIFAR-10 (Li et al., 2017). Figure 1 illustrates the test accuracy of various algorithms versus the number of communication rounds on the EMNIST dataset, with subfigure 1a representing i.i.d. data and subfigure 1b depicting non-i.i.d. data. Figure 2 compares the test accuracy of various algorithms versus the learning rate on the EMNIST dataset.
Researcher Affiliation Academia Wenjing Yan1, Kai Zhang2, Xiaolu Wang3, Xuanyu Cao4 1The Chinese University of Hong Kong 2The Hong Kong University of Science and Technology 3East China Normal University 4Washington State University {wjyan}@ie.cuhk.edu.hk, {kzhangbn}@connect.ust.hk, {xiaoluwang}@sei.ecnu.edu.cn, {xuanyu.cao}@wsu.edu
Pseudocode Yes Algorithm 1 PAda MFed: A Problem-Parameter-Agnostic Algorithm for Nonconvex FL ... Algorithm 2 PAda MFed-VR: PAda MFed with Variance Reduction
Open Source Code No The paper does not contain any explicit statement about open-sourcing the code, nor does it provide a link to a code repository. The text mentions experimental validation and numerical experiments but does not include any specific information regarding the availability of the implementation code.
Open Datasets Yes In this section, we present experiments on two real-world datasets: EMNIST (Cohen et al., 2017) and CIFAR-10 (Li et et al., 2017).
Dataset Splits No The paper mentions distributing data across clients for i.i.d. and non-i.i.d. scenarios using uniform random distribution or Dirichlet distribution. It also states the number of clients and participating clients. However, it does not specify the explicit training, validation, or test dataset splits (e.g., percentages or sample counts) for reproducibility.
Hardware Specification No The paper does not provide any specific hardware details such as CPU/GPU models, memory specifications, or cloud computing instances used for running the experiments. It describes the experimental setup in terms of datasets, model architectures, and client participation but omits hardware information.
Software Dependencies No The paper mentions using a 'convolutional neural network (CNN)' and 'Res Net-18 architecture' which implies the use of deep learning frameworks, but it does not specify any software dependencies (e.g., PyTorch, TensorFlow) along with their version numbers.
Experiment Setup Yes The experimental framework involves 100 distributed clients with 10 clients participating randomly in each training round. We employ a convolutional neural network (CNN) with three convolutional layers and two fully connected layers for the EMNIST dataset, and a Res Net-18 architecture for CIFAR-10. All algorithms were evaluated over 400 communication rounds to ensure a fair comparison. The hyperparameters of all baselines, including learning rates, are optimized through comprehensive grid search. The stepsizes for our algorithms, PAda MFed and PAda MFed-VR, are determined based on the theoretical guidance provided in Theorem 1 and Theorem 2, respectively.