Wednesday Notes

Agent Orchestration  & Federative Learning

Project: Build an agent-based federative learning system that can be used assess a scientists analysis. The output presentation is without data centralisation (from adjacent biobanks). 

Core questions:
How data is located and localised from multiple sources?
What does the data output look like with segmentation completed by a federated agent?

Workflow ideas:

Global server prompted by user task -> root request to the clients with data to support the task -> local agent observes the task -> global agent can observe the encapsulated data now too -> knowledge annotations prompt response to server 

A for-loop can be used to re-prompt the user to enter for more input information if the local sourcing to global agent output is unclear.

Prompt response -> server-agent segmentation (validation codes for local client data validation) -> global server agent will describe the data contract -> if success -> local agent conferred and validates data through folds -> server agent can begin coding now on the local data

Input: To Begin Task In 
Processing: Data Function/Algorithm that standardises and cleans data for the global model
Output: End Analysis Results/Descriptions Out

Feeback after Workflow presentation:
Algorithms do not need to be complicated. Fx XGBoost, RandomForest.