20 datasets across vision, NLP, and speech categories for federated learning research. The platform provides a standardized environment for evaluating both the statistical accuracy and system efficiency of distributed learning models.
Use Cases
- Train image classification models using the image and label columns in the FEMNIST dataset
- Simulate communication overhead using the network_bandwidth and latency features from device traces
- Benchmark NLP models using the text and client_id fields from the Reddit dataset
Strengths
- Includes 20 datasets covering vision, NLP, and speech tasks
- Contains system traces for 2,000+ mobile devices including battery and network profiles
- Provides standardized client_id mapping for non-IID data partitioning