| AIH-HSP | AI Workflows | AI Modules |
AIH-HSP
MPAI-AIH is an ecosystem of:
- Multiple Front End-oriented PAAIs (FE-PAAI) that:
- Capture Health Data from End Users.
- Process Health Data to produce AIH Data.
- Learn while processing Health Data.
- Sends licensed AIH Data to the BE-PAAI.
- Send trained Neural Networks to the Back End for Federated Learning.
- Download new versions of Neural Networks produced by Federated Learning.
- One Back End-oriented PAAI (BE-PAAI) that:
- Receives licensed AIH Data from FE-PAAIs.
- Stores licences as smart contracts in a blockchain.
- Enables Third-Party Users to process AIH Data based on licences.
- Requests trained Neural Networks from FE-PAAIs.
- Produces new versions of Neural Networks by Federated Learning.
- Uploads new versions of Neural Networks to MPAI Store for FE-PAAIs to download.
Figure 1 depicts the AIH-HSP ecosystem.
Figure 1 – General Model if AIH-HSP V1.0
1 AI Workflows
1.1 Health Back End
HSP-HBE is a PAAI composed of collaborating PAAIs.
Figure 1 – Reference Model of the Health Back End (AIH-HBE) AIW
The following links analyse the AI Modules:
AIH-ADT Auditing
AIH-AAC Authentication and Access Control
AIH-BDP HBE Data Processing
AIH-DSA Data Storage and Access
AIH-DIA De-Identification and Anonymisation
AIH-HFF Health Federated Learning
1.2 Health Front End
HSP-HFE is a PAAI composed of collaborating PAAIs.
Figure 1 – Reference Model of the Health Front End (AIH-HFE) AIW
The following links analyse the AI Modules:
AIH-ARA Anomaly and Risk Alert
AIH-HDM AIH Data Multiplexing
AIH-FDP HFE Data Processing
2 AI Modules
2.1 Anomaly and Risk Alert
2.2 Auditing
2.3 Authentication and Access Control
2.4 Data Storage and Access
2.5 De-Identification and Anonymisation
2.6 HBE Data Processing
AIH-BDP is a PAAI
| Receiving | AIH Data | From HFEs or internal Secure Storage |
| Processing Request | From processes inside the HBE or Third-Party Users. | |
| NN Request | From Health Back End | |
| NN Model | As trained processing End User’s AIH Data. | |
| Producing | AIH Data | To internal Secure Storage and to requesting Third-Party Users |
| NN Model | Resulting from Federated Learning to MPAI Store. | |
| NN Request | To MPAI Store |
An HBE is composed of an AI Framework where AIWs/AIMs downloaded from the MPAI Store are executed to process AIH Data
- At the request of:
- Internal processes of the HBE.
- Third-Party Users.
- Verified for processability by requesting a licence to the blockchain.
AIH Data have been
1.2.7 HFE Data Processing
AIH-FDP is a PAAI
| Receiving | Processing Request | From End-User |
| AIH Data | Requested to and received from Secure Storage | |
| NN Request | From Health Back End for Federated Learning | |
| Neural Network | From MPAI Store | |
| Producing | AIH Data | Stored at Secure Storage |
| NN Model | Sent to Health Back End | |
| NN Request | To MPAI Store |
An HFE is composed of an AI Framework where AIWs/AIMs downloaded from the MPAI Store are executed to process AIH Data that
- Have been acquired and stored in the Secure Storage.
- Are read from the Secure Store.
AIMs processing AIH Data in the Health Front End are designed to learn from the AIH Data they process, i.e., the Neural Networks processing the AIH Data gradually change their parameters. From time to time, the Health Back End collects the AIMs of a certain type from all Health Front Ends, produces a new version of the AIM by performing Federated Learning and uploads the resulting AIM to the MPAI Store. Health Front Ends will then download the new version of that type of AIM.
2.8 Health Data Multiplexing
2.9 Health Federated Learning
From time to time, the AIH-HFL PAAI requests AIH-HFEs to share the Neural Networks used in specific a AIM.
| Receiving | NN Model | From HFEs |
| Extracting | Knowledge | From received NN Model using Federated Learning. |
| Producing | NN Request | To HFEs |
| Uploading | NN Models | To MPAI Store frof HFEs to downloa. |
The Health Federated Learning AIM performs Learning Level Operations.
