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].
Multimodal Fusion Using Multi-View Domains for Data Heterogeneity in Federated Learning
Authors: Min Gao, Haifeng Zheng, Xinxin Feng, Ran Tao
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive simulations demonstrate the superiority of the proposed framework, which exhibits significant merits in tackling model degeneration caused by data heterogeneity and enhancing modalitybased generalization for heterogeneous scenarios. We evaluate the proposed framework using two real-world multimodal datasets. The results demonstrate the superiority of our framework |
| Researcher Affiliation | Academia | 1College of Physics and Information Engineering, Fuzhou University, Fuzhou, China 2School of Information and Electronics, Beijing Institute of Technology, Beijing, China |
| Pseudocode | No | The paper describes the proposed methods (GLA and LAM) in detail using mathematical formulations and textual explanations, but it does not present them in a structured pseudocode block or algorithm format. |
| Open Source Code | No | The paper does not contain an unambiguous statement about releasing source code, nor does it provide a link to a code repository or mention code in supplementary materials. |
| Open Datasets | Yes | We evaluate the proposed methods on two typical multimodal datasets named FLASH (Salehi et al. 2022; Ouyang et al. 2023) and UTD-MHAD (Chen, Jafari, and Kehtarnavaz 2015; Ouyang et al. 2022) |
| Dataset Splits | No | The paper defines a 'global training dataset' and a 'testing dataset' conceptually for the FL framework. It also describes how heterogeneity is generated for clients using Dirichlet distribution (Dir(β)) and controlling modality ratios (Θ). However, it does not provide specific global percentages (e.g., 80% train, 20% test) or sample counts for the overall FLASH or UTD-MHAD datasets for replication. |
| Hardware Specification | No | The paper mentions 'heterogeneous IoT nodes equipped with various modal sensors' as part of the conceptual framework, but it does not specify any particular hardware (e.g., GPU/CPU models, memory) used to run the experiments or train the models. |
| Software Dependencies | No | The paper does not provide any specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions or other libraries). |
| Experiment Setup | Yes | We simulate modality-based data heterogeneity on the naturally heterogeneous FLASH dataset containing 210 heterogeneous nodes, and both categories-based and modality-based data heterogeneity on the UTD-MHAD dataset. For category-based heterogeneous setting, following the previous works (Huang et al. 2021; Li, He, and Song 2021), we generate non-i.i.d. data partition with Dirichlet distribution, denoted as Dir(β), where the smaller β controls more heterogeneous setting (0.1 by default). As for the modality-based heterogeneous setting, similar to previous works (Ouyang et al. 2023), we apply Θ = {αm|m M} to control the ratios of nodes possessing m modalities, where PM m=1 αm = 1. ε and α are hyper-parameters to prevent the denominator from zero and control the strength of pushing the decision boundary, respectively. They are set as small values in our experiments. |