LEAF
收藏资源简介:
LEAF是由卡内基梅隆大学创建的一个模块化基准框架,专注于联邦学习环境。该框架包含六个数据集,分别是FEMNIST、Sent140、Shakespeare、CelebA、Reddit和一个合成数据集。这些数据集的特点是数据自然地按设备/用户生成,涉及从数千到数百万的设备网络,且数据点在设备间分布不均。LEAF的创建旨在为联邦学习、元学习和多任务学习等领域提供真实世界的基准,解决数据异构性、系统规模和隐私安全等挑战。
LEAF is a modular benchmark framework developed by Carnegie Mellon University, focusing on federated learning environments. It comprises six datasets: FEMNIST, Sent140, Shakespeare, CelebA, Reddit, and a synthetic dataset. These datasets feature data naturally generated per device or user, involve device networks ranging from thousands to millions of nodes, and present non-uniform distributions of data points across devices. LEAF was created to provide real-world benchmarks for domains including federated learning, meta-learning, and multi-task learning, addressing core challenges such as data heterogeneity, system scalability, and privacy and security issues.

- 1LEAF: A Benchmark for Federated Settings卡内基梅隆大学 · 2019年



