Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
FedSoft: Soft Clustered Federated Learning with Proximal Local Updating
Authors: Yichen Ruan, Carlee Joe-Wong8124-8131
AAAI 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we verify the effectiveness of Fed Soft with two base datasets under various mixture patterns. For all experiments, we use N = 100 clients, and the number of samples in each client nk is chosen uniformly at random from 100 to 200. For ease of demonstration, for every base dataset, we ο¬rst investigate the mixture of S = 2 distributions and then increase S. and Table 2 compares Fed Soft with the baselines. Not only does Fed Soft produce more accurate cluster and local models, but it also achieves better balance between the two trained centers. |
| Researcher Affiliation | Academia | Yichen Ruan, Carlee Joe-Wong Carnegie Mellon University EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1: Fed Soft |
| Open Source Code | No | The paper does not contain an explicit statement about releasing its source code or a link to a code repository. |
| Open Datasets | Yes | We use three datasets to generate the various distributions: Synthetic, EMNIST and CIFAR-10. |
| Dataset Splits | No | The paper states it uses 'test accuracy/error on holdout datasets' and 'evaluate their accuracy/error on local training datasets' but does not provide specific percentages or counts for train/validation/test splits of the datasets used (Synthetic, EMNIST, CIFAR-10). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers. |
| Experiment Setup | Yes | Unless otherwise noted, we choose Fed Soft s estimation interval Ο = 2, client selection size K = 60, counter smoother Ο = 1e-4, and all experiments are run until both cluster and client models have fully converged. All models are randomly initialized with the Xavier normal (Glorot and Bengio 2010) initializer without pre-training, so that the association among clients, centers, and cluster distributions is built automatically during the training process. |