H-nobs: Achieving Certified Fairness and Robustness in Distributed Learning on Heterogeneous Datasets

Authors: Guanqiang Zhou, Ping Xu, Yue Wang, Zhi Tian

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we empirically evaluate the performance of our algorithm for a classification task using the logistic regression model on the Spambase dataset.
Researcher Affiliation Academia Guanqiang Zhou George Mason University gzhou4@gmu.edu Ping Xu University of Texas Rio Grande Valley ping.t.xu@utrgv.edu Yue Wang Georgia State University ywang182@gsu.edu Zhi Tian George Mason University ztian1@gmu.edu
Pseudocode Yes Algorithm 1 Norm-Based Screening, Algorithm 2 H-nobs: Fair & Byzantine-Robust Distributed Gradient Descent
Open Source Code No The paper does not provide an explicit statement about open-sourcing the code for its proposed methodology or a link to a code repository.
Open Datasets Yes using the logistic regression model on the Spambase dataset [21]. We assign 2/3 of the 4601 total samples for training and the other 1/3 for testing. The results for these two datasets are presented in Appendix E. Table 3: The performance of H-nobs on the Law School dataset (with no attack). Table 4: The performance of H-nobs on the Credit Card Client dataset (with no attack).
Dataset Splits No We assign 2/3 of the 4601 total samples for training and the other 1/3 for testing. The paper explicitly mentions training and testing splits, but does not specify a separate validation split or its size/proportion.
Hardware Specification No The paper does not specify any hardware details such as GPU models, CPU types, or cloud computing resources used for the experiments.
Software Dependencies No The paper mentions using a 'logistic regression model' but does not specify any software libraries or their version numbers, such as Python, PyTorch, TensorFlow, or scikit-learn.
Experiment Setup Yes In Figure 1, we plot the performance curves of model accuracy for the five considered aggregation measures... using learning rate η = 1 and number of iterations T = 300. In Table 1, we document both the model accuracy and the variance of local accuracies (the metric of model fairness) using different values of q and screening percentage β. The learning rate η and number of iterations T for each q value are carefully selected to ensure fast and stable convergence. Finally, we compare H-nobs... using learning rate η = 0.5 and number of iterations T = 1000.