SAGDA: Achieving $\mathcal{O}(\epsilon^{-2})$ Communication Complexity in Federated Min-Max Learning

Authors: Haibo Yang, Zhuqing Liu, Xin Zhang, Jia Liu

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we conduct numerical experiments using two machine learning problems (Logistic Regression and AUC Maximization) to verify our theoretical results for SAGDA as well as FSGDA.
Researcher Affiliation Academia Haibo Yang Dept. of ECE The Ohio State University Columbus, OH 43210 yang.5952@osu.edu; Zhuqing Liu Dept. of ECE The Ohio State University Columbus, OH 43210 liu.9384@osu.edu; Xin Zhang Dept. of Statistics Iowa State University Ames, IA 50010 xinzhang@iastate.edu; Jia Liu Dept. of ECE The Ohio State University Columbus, OH 43210 liu@ece.osu.edu
Pseudocode Yes Algorithm 1 The Stochastic Averaging Gradient Descent Ascent (SAGDA) Algorithm.; Algorithm 2 Federated Stochastic Gradient Descent Ascent (FSGDA) Algorithm.
Open Source Code No Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] See detailed instructions in Section 4 and Appendix, but codes are private.
Open Datasets Yes We test the convergence performance of our algorithms using the a9a dataset [40] and MNIST [41] from LIBSVM repository.
Dataset Splits No The paper describes how data is distributed across clients for federated learning, and how a subset of the MNIST dataset was created, but not standard train/validation/test splits of the overall dataset used for evaluation.
Hardware Specification No Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [No]
Software Dependencies No The paper does not provide specific software dependencies with version numbers for reproducibility.
Experiment Setup Yes The learning rates are chosen as ηx,l = ηy,l = 10 2, ηx,g = ηy,g = 2, local updates K = 10.