Fairness in Federated Learning via Core-Stability
Authors: Bhaskar Ray Chaudhury, Linyi Li, Mintong Kang, Bo Li, Ruta Mehta
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we empirically validate our analysis on two real-world datasets, and we show that Core Fed achieves higher core-stable fairness than Fed Avg while maintaining similar accuracy. We evaluate our fair ML method Core Fed and baseline Fed Avg [21] on three datasets (Adult, MNIST and CIFAR-10) on linear model and deep neural networks. |
| Researcher Affiliation | Academia | Bhaskar Ray Chaudhury Linyi Li Mintong Kang Bo Li Ruta Mehta University of Illinois at Urbana Champaign |
| Pseudocode | Yes | We call our Algorithm as Core Fed (Fully outlined in Algorithm 1 in the appendix). |
| Open Source Code | Yes | 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] |
| Open Datasets | Yes | We evaluate our algorithm Core Fed on Adult [2], MNIST [17] and CIFAR-10 [16] datasets. |
| Dataset Splits | No | The paper mentions data construction for non-IID settings but does not explicitly provide specific percentages, sample counts, or methodology for training, validation, or test dataset splits. |
| Hardware Specification | Yes | All experiments are conducted on a 1080 Ti GPU. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers (e.g., library or solver names with version numbers like Python 3.8, CPLEX 12.4) needed to replicate the experiment. |
| Experiment Setup | Yes | For Adult dataset, the utitility is selected as M ℓlog where ℓlog is the logistic loss. For CIFAR-10 and MNIST, we use cross entropy loss ℓce as the training loss with utility U becomes M ℓce. M is set to be 3.0, 1.0 and 3.0 for Adult, MNIST, and CIFAR-10, respectively, based on statistical analysis during training. We use a CNN, which has two 5x5 convolution layers followed by 2x2 max pooling and two fully connected layer with Re LU activation for MNIST and CIFAR-10. |