Federated Learning
A planned capability to train AI models across the FractalAI network without sharing raw data. Privacy-preserving, decentralized machine learning — in design, not yet operational.
Privacy-Preserving (planned)
The design keeps raw data on the node — only model gradients would be shared. Privacy-by-design is a core goal.
Distributed Training (planned)
Multiple nodes would train FANE models in parallel, with aggregated updates improving the global model.
Cryptographic Verification (planned)
Gradient submissions would be signed with Dilithium-3 (ML-DSA-65) for tamper-evident contribution tracking.
Contribution Rewards (planned)
A future reward mechanism could compensate nodes that contribute quality gradients. Not implemented yet.
How It Works
Global Model Distribution
Current FANE model weights would be distributed to participating nodes.
Local Training
Each node would train on its local data. Raw data never leaves the node.
Gradient Submission
Computed gradients would be signed and submitted for aggregation.
Secure Aggregation
Gradients would be aggregated using secure multi-party computation.
Model Update
The global FANE model would be updated, and all nodes receive the improved model.
Follow the Roadmap
Federated learning is in design. Node participation and any contribution rewards are not available yet — follow the docs for updates.
Read the Docs