Daml Code Assistant #10 - Milestone 4: Federated learning system

Milestone 4: Federated learning system

• Estimated Delivery: +240 Days from CIP Approval
• Focus: Building an opt-in federated-learning system that lets self-hosting organizations contribute to continuous model improvement without exposing their Daml code.
• Deliverables / Value Metrics: Deliver the federated-learning system supporting multiple participating organizations: automated on-premise gathering of local training data from development signals, local training that produces weight updates shared without exposing code, central aggregation that produces improved models, participant authentication, robustness to low-quality or malicious updates, the opt-in mechanism granting participants discounted access to improved models, privacy guarantees that no code leaves participants' premises, and packaging/documentation for self-hosted deployment; demonstrated by an end-to-end federated round across at least two controlled client deployments holding Daml data withheld from central training, in which the aggregated model measurably improves on an evaluation reflecting that withheld data, confirming the loop transfers learning from local data without that data being centrally pooled.

Milestone | Description | Payment -- | -- | -- Milestone 4 | Federated Learning System | 3,312,126 CC (equivalent to 546,501 USD) upon acceptance of the federated-learning system

_Originally posted by @pedrodneves in https://github.com/canton-foundation/canton-dev-fund/issues/10#issuecomment-4979795624_