Federated Model Interface (FMI)
The FMI is the high-level API that ties all the components together for Federated Learning. It serves as the gateway for all external model communication.
The Pipeline
When a packet arrives from a peer, it goes through 4 stages:
-
Validation
- Checks cryptographic signature.
- Verifies schema version.
- Checks sender Trust Score.
-
Routing
- Assigns the packet to the correct Peer Cohort.
- Filters based on domain relevance.
-
Aggregation
- Combines the parameters with local knowledge.
- Uses FedAvg or Adaptive Weighted Averaging.
-
Emission
- Notifies subscribers (Engine/Meta-Learner) of the new state.
DRG Integration
The FMI automatically degrades performance (gracefully) when resources are tight:
- Low Bandwidth? -> Switch to
Q8(8-bit) quantization. - High Latency? -> Suspend Proof-of-Path (POP) verification.
- High VRAM? -> Defer aggregation until memory clears.
- Resources OK? -> Restore full precision (
FP32).