Federated Learning with Convex Global and Local Constraints

Authors: Chuan He, Le Peng, Ju Sun

TMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our numerical experiments show the effectiveness of our algorithm in performing Neyman-Pearson classification and fairness-aware learning with nonconvex constraints, in an FL setting.
Researcher Affiliation Academia Chuan He EMAIL Department of Computer Science and Engineering, University of Minnesota Le Peng EMAIL Department of Computer Science and Engineering, University of Minnesota Ju Sun EMAIL Department of Computer Science and Engineering, University of Minnesota
Pseudocode Yes Algorithm 1 A proximal AL based FL algorithm for solving Eq. (5) Algorithm 2 An inexact ADMM based FL algorithm for solving Eq. (16)
Open Source Code Yes The code to implement the proposed algorithm on these numerical examples is available at https://github.com/PL97/Constr_FL.
Open Datasets Yes We consider three real-world datasets, namely breast-cancer-wisc , adult-a , and monks-1 , from the UCI repository2 and described in Appendix E.1. For the real-world dataset, we consider adult-b 3: each sample in this dataset has 39 features and one binary label. This dataset can be found in https://github.com/heyaudace/ml-bias-fairness/tree/master/data/adult.
Dataset Splits Yes To simulate the FL setting, we divide each dataset into n folds, mimicking local clients, each holding the same amount of data with equal ratios of the two classes. To simulate the FL setting, we divide the 22, 654 training samples from the adult-b dataset into n folds and distribute them to n local clients. The central server holds the 5, 659 test samples from the adultb dataset.
Hardware Specification Yes All experiments are carried out on a Windows system with an AMD EPYC 7763 64-core processor
Software Dependencies No All experiments are carried out on a Windows system with an AMD EPYC 7763 64-core processor, and all algorithms are implemented in Python. The paper mentions Python but does not specify a version number or any other software dependencies with version numbers.
Experiment Setup Yes We set the other parameters for Algorithm 1 and c Prox-AL as µ0 i = (0, . . . , 0)T 0 i n, s = 0.001 and β = 300. We also set ρi = 0.01 1 i n for Algorithm 2. We set the other parameters for Algorithm 1 and c Prox-AL as µ0 i = (0, . . . , 0)T 0 i n, s = 0.001 and β = 10. We also set ρi = 108 1 i n for Algorithm 2.