Train
Train is the Compex workflow for governed model training on protected data. It supports training patterns where the work moves toward the data under policy, approval, audit, and evidence controls instead of requiring default raw-data handover.
What Train supports
Train is not limited to federated learning. Depending on the use case and risk model, a Train pilot can use:
- clean-room training;
- owner-side training;
- internal governed training;
- cross-organizational training;
- federated learning where needed;
- future confidential execution patterns as the roadmap matures.
How it works
Define data and scope
The data owner identifies the protected dataset, allowed training purpose, users, retention expectations, and output rules.
Approve the workload
Technical and review teams inspect the workload package before it runs. The approval is recorded as part of the evidence trail.
Run under governance
Compex executes the approved training workflow within the agreed boundary and records relevant events.
Review outputs
Outputs such as model artifacts, metrics, or evaluation reports are reviewed according to policy before release.
What this means
For business readers, Train creates a way to test whether protected data can improve a model without starting from a raw-data transfer negotiation.
Current status
Train is pilot-ready for scoped evaluations where training data, workload package, output type, and review roles are defined upfront.
Limitations
Compex does not guarantee that a trained model is non-sensitive, legally approved, or safe for all downstream use. Output review and legal assessment remain part of the pilot process.
Compare Train with Analyze and Serve
Use the comparison guide if you are deciding whether a use case needs training, computation, or a governed service.