The intersection of simulation and artificial intelligence (AI) was the subject of AnyLogic’s AI workshop at the 2020 Winter Simulation Conference. This intersection is benefiting the technologies themselves and providing businesses with new insights and decision support.
During the workshop, AnyLogic’s AI Program Lead Dr. Arash Mahdavi and Simulation Specialist Tyler Wolfe-Adam presented simulation and AI concepts and technologies in three main parts:
- the differences between machine learning (ML) and simulation, including their pros and cons,
- how ML and simulation complement each other,
- a demonstration of AnyLogic as a simulation platform for applied AI.
After introducing the technologies and comparing them, the workshop details and demonstrates how the two disciplines can benefit each other. The benefits of combining simulation and AI come from using them for three main purposes:
- Synthetic data generation
- Learning environments
- Testbeds for trained AI
How useful are those combinations to the real world and how easy are they to implement?
For synthetic data generation, the workshop presents several use cases and shows how AnyLogic’s database connectivity, Cloud API (JS, Python, Java), and the upcoming ALPyne Python connector provide flexible access to synthetic data. Also, with the release of AnyLogic 8.7, a new Reinforcement Learning experiment is available. The workshop shows how you can use this new experiment to directly connect to automated AI development environments.
In the case of learning environments, simulation models are a place for reinforcement learning (RL). This means a machine learning AI agent can learn what to do: how to map situations to actions for the best outcome. In effect, the AI agent replaces a human in a system – completing a feedback loop of actions and changes.
For the last of the three main areas, the workshop shows how simulations can function as a testbed for trained AI. There are six cases where a simulation hosting an ML agent can be useful:
- Using ML models as alternative model inputs,
- Using ML models to approximate component behavior in a simulated system,
- Inserting operational ML into a simulated environment for a more accurate simulation,
- Testing an AI solution before deployment,
- Visualizing the behavior of code/math,
- Testing an RL policy
Again, methods for integrating AI and simulation accompany each section and make use of AnyLogic’s versatility. There are options for connecting to popular ML/AutoML platforms, importing AI solutions natively into the simulation environment, and for using Pypline to access AI solutions in Python.
Watch the workshop recording to learn more about the technology and concepts involved in combining simulation and AI.