Steering Architecture

There are four kinds of architecture depending on the setup for ABM - Data - ML integration.

Architecture A:

• There is no empirical data, but agents execute actions that lead to feedback (Reinforcement Learning).
• The data is collected and fed to the ML Algorithm that starts to learn how to predict correctly.
• The result of the ML steer agent behaviour.

Architecture B:

• Empirical data train the ML algorithm.
• The ABM model generates new data that is sent to the trained ML
• The ML algorithm is retrained, and the result is sent back to ABM

Architecure C:

• Combines A and B
• In the initialization phase the ML algorithm is half trained and its training will continue during the simulation.

Architecture D:

Is the same process as B but no data is available at the start so is generated by the ABM model itself.