LEAF

A Benchmark for Federated Settings

LEAF is a benchmarking framework for learning in federated settings, with applications including federated learning, multi-task learning, meta-learning, and on-device learning. Future releases will include additional tasks and datasets. Please contact Sebastian Caldas with questions or to contribute to the benchmark.

Get Started GitHub

Statistics
3,550 users
80,5263 samples (total)
226.83 samples per user (mean)
num_samples (std): 88.94
num_samples (std/mean): 0.39

Statistics
1,129 users
4,226,15 samples (total)
3,743.2 samples per user (mean)
num_samples (std): 6,212.26
num_samples (std/mean): 1.65

Statistics
660,120 users
1,600,498 samples (total)
2.42 samples per user (mean)
num_samples (std): 4.71
num_samples (std/mean): 1.94

Statistics
9,343 users
200,288 samples (total)
21.44 samples per user (mean)
num_samples (std): 7.63
num_samples (std/mean): 0.36

See the original Large-scale CelebFaces Attributes Dataset

Default Distribution (parameters are customizable)
Statistics
1,000 users
107,553 samples (total)
107.55 samples per user (mean)
num_samples (std): 213.22
num_samples (std/mean): 1.98

Statistics
1,660,820 users
56,587,343 samples (total)
34.07 samples per user (mean)
num_samples (std): 62.9
num_samples (std/mean): 1.84

Citation

Please cite the following if you use LEAF:

LEAF: A Benchmark for Federated Settings
Sebastian Caldas, Sai Meher Karthik Duddu, Peter Wu, Tian Li,
Jakub Konečný, H. Brendan McMahan, Virginia Smith, and Ameet Talwalkar.
Workshop on Federated Learning for Data Privacy and Confidentiality (2019).