Structured Federated Learning through Clustered Additive Modeling
Authors: Jie Ma, Tianyi Zhou, Guodong Long, Jing Jiang, Chengqi Zhang
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Experiments, Table 1: Test results (mean std) in cluster-wise non-IID settings on Fashion-MNIST & CIFAR-10., Figure 1: Cluster sizes during IFCA vs. IFCA+CAM in client/cluster-wise non-IID settings on CIFAR-10. Legend: cluster ID (cluster size) in the last round. CAM effectively mitigates clustering collapse/imbalance. |
| Researcher Affiliation | Academia | 1Australian Artificial Intelligence Institute, FEIT, University of Technology Sydney 2University of Maryland |
| Pseudocode | Yes | Algorithm 1: Fed-CAM |
| Open Source Code | No | The paper does not provide any statement or link indicating the public availability of the source code for the described methodology. |
| Open Datasets | Yes | Fashion-MNIST [29] includes 70,000 labeled fashion images (28 28 grayscale) in 10 classes..., CIFAR-10 [30] consists of 60,000 images (32 32 color) in 10 classes..., Path MNIST and Tissue MNIST from the Med MNIST [28] |
| Dataset Splits | No | The paper mentions training and testing on datasets but does not explicitly provide specific percentages or counts for train/validation/test splits, nor does it define a validation set split. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or processor types) used for running its experiments. |
| Software Dependencies | No | The paper mentions using SGD and CNN but does not provide specific version numbers for software libraries, frameworks, or programming languages used in the experiments. |
| Experiment Setup | Yes | For optimization, we employ SGD with a learning rate of 0.001 and momentum of 0.9 to train the model, and the batch size is 32. We conduct 100 global communication rounds in the FL system, including 30 warm-up rounds if applicable. Each communication involves 10 local steps. For the clustering process of Fe SEM-CAM, we measure distance on the flattened parameters of the fully-connected layers, and use K-Means as the clustering algorithm. The coefficient λ is chosen from 0.001, 0.01, 0.1 based on the best performance. |