The DEPA training framework supports a broad class of data sharing scenarios, including
- Data sharing for societal benefit e.g., sharing patient health data with research institutions for training better diagnostic models.
- Data sharing for commercial benefit of the TDC e.g., sharing travel data with a partner organization to help make personalized recommendations.
- Data sharing for commercial benefit of the TDP e.g., monetizing carefully curated and labelled datasets.
- Data sharing for mutual benefit of TDPs e.g., sharing financial transaction data for training an accurate fraud detection model.
TDC -> TDP Payments
In some scenarios, TDPs MAY expect payments from TDCs for the use of their datasets, either for covering the costs of curating and hosting datasets, or for commercial gains. Payment terms can include one-time payments or pay-per-use.
The DEPA training framework supports a transparent, automated and flexible mechanism for TDPs to receive payments from TDCs.
- Payment terms declaration: TDP declare the terms of payments in their signed dataset references registered with the contract service. The terms SHOULD be fair, reasonably and non-discriminatory i.e., apply equally to all registered TDCs.
- Contract signing: TDCs indicate agreement with the terms by signing contracts, and registering the contracts with a contract service. TDPs and TDCs receive signed receipts from the contract service on registration; these receipts may be used to prove registration and validity of the contract.
- Notifications: CCRs generate notifications to the contract service during the training workflow e.g., after fetching data encryption keys and before releasing the trained model to the TDC.
- Payments: Contract services trigger payments from TDCs to TDPs using external payment systems.
TDP -> DP Payments
In some scenarios, TDPs MAY compensate data principals for contributing to their datasets, either voluntarily or as required by the SRO.
The DEPA Training Framework currently does not define the nature of this compensation or support automated payments. However, it encourages TDPs to explore appropriate ways to compensate data principals in a timely manner, ensuring that compensation does not create perverse incentives.
TDC/TDP -> CCR Provider Payments
In scenarios where the CCR is being hosted by a CCR provider on behalf of the TDC or TDP, the CCR provider may be paid for the resources (compute, memory, etc.) used during training. The framework currently does not define the nature of this compensation or support automated payments.