FedMultimodal
收藏资源简介:
FedMultimodal是由南加州大学开发的一个多模态联邦学习基准数据集,涵盖了五个代表性的多模态应用场景,包括情感识别、多媒体动作识别、人类活动识别、医疗健康和社交媒体。该数据集包含十个常用数据集,总计涉及八种独特的模态。FedMultimodal提供了一个系统的联邦学习流程,从数据分割和特征提取到联邦学习基准算法和模型评估,支持端到端的建模框架。此外,该数据集还提供了一个标准化方法来评估联邦学习对抗三种常见数据损坏的鲁棒性,包括模态缺失、标签缺失和错误标签。
FedMultimodal is a multimodal federated learning benchmark dataset developed by the University of Southern California. It encompasses five representative multimodal application scenarios, including emotion recognition, multimedia action recognition, human activity recognition, healthcare, and social media. This dataset incorporates ten commonly used datasets, involving a total of eight distinct modalities. FedMultimodal provides a systematic federated learning workflow ranging from data partitioning and feature extraction to federated learning benchmark algorithms and model evaluation, supporting end-to-end modeling frameworks. Additionally, this dataset offers a standardized method to evaluate the robustness of federated learning against three common data corruptions: missing modality, missing label, and incorrect label.




