Federated Generative Model on Multi-Source Heterogeneous Data in IoT
Authors: Zuobin Xiong, Wei Li, Zhipeng Cai
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on a simulated dataset and multiple real datasets are conducted to evaluate the data generation performance of our proposed generative models through comparison with the state-of-the-arts. |
| Researcher Affiliation | Academia | Department of Computer Science, Georgia State University, 25 Park Place, Atlanta, Georgia, 30303 USA zxiong2@student.gsu.edu, {wli28, zcai}@gsu.edu |
| Pseudocode | No | The paper describes the training processes in text but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statement about the release of source code for the described methodology or a link to a code repository. |
| Open Datasets | Yes | For the feature related scenario, we choose two types of data in the experiments: (i) simulated gaussian mixed data, where the datasets of all local communities have the same feature (i.e., variance) but different mean values; and (ii) MNIST dataset and Fashion-MNIST dataset, where both datasets have the same features but different class labels. For the label related scenario, we select (i) MNIST dataset and inverse MNIST dataset; (ii) sketchphoto dataset (Zhu et al. 2016), each of which has the same labels or semantic information but different domains. |
| Dataset Splits | No | Data Distribution. Both the i.i.d. data distribution and the non-i.i.d. data distribution are considered for the feature related and the label related scenarios in our experiments. Accordingly, by combining the two types of data distribution and the two scenarios, we have the following four types of settings for evaluation: (i) feature related scenario with i.i.d. data distribution, (ii) feature related scenario with non-i.i.d. data distribution, (iii) label related scenario with i.i.d. data distribution, and (iv) label related scenario with non-i.i.d. data distribution. |
| Hardware Specification | No | The network environment is simulated on one server, where each community has 1 edge server and 5 Io T devices. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions, or other libraries). |
| Experiment Setup | Yes | The global aggregation is performed by Igl iterations, each of which contains Ilo iterations of local community training. and Therefore, Igl and Ilo can be set by considering the trade-off between training result and training time. and Particularly, λ1, λ2 (0, 1] are two pre-determined hyperparameters used as the weights of loss functions. |