Automated systems have reached their limits. As a result, companies wanting to further enhance business processes are turning to artificial intelligence (AI) technologies like machine learning (ML).
In an AnyLogic workshop, Microsoft Autonomous Systems Principal Program Manager Kence Anderson explored the advanced decision-making possibilities of ML and showed how Microsoft’s machine teaching concept is achieving faster training times.
The easily accessible workshop session provides a high-level overview of AI’s state of the art with examples from Microsoft and DeepMind research, as well as illustrative Karate Kid analogies.
Low-code easy access
The technologies described in the session have moved beyond research and are now easily accessible and powerful business tools. For example, Microsoft’s Project Bonsai is a low-code platform for AI automation of existing processes and equipment without the need for data scientists.
Project Bonsai exemplifies the modern approaches to machine learning described in the session. Training takes place in the safety of simulated environments that do not suffer from downtime situations which may occur in the real world. Then, once trained, AI agents can be deployed for real-time control of a process or system.
The session shows how using AI makes it possible to consider all available data and to flexibly respond to new scenarios with new behavior. This is in contrast to other approaches, such as expert systems, optimization, and the Monte Carlo method, which can be fixed in their behavior or restricted in the data they consider.
Furthermore, a key difference of AI-based systems from alternative system control approaches is how AI can optimize for better results later. This means that an AI agent does not simply take the best option at each moment but will take an approach that delivers better results overall.
Problems training AI
The workshop session addresses the machine learning problem of resource intensive training.
Although machine learning can be powerful tool for improving system control and efficiency, the training needed to achieve good results can be vast. Consequently, there can be a need for powerful computers and the electricity to run them, possibly for long periods of time.
The solution proposed by Anderson, is not just to use cloud computing but for modularized AI brains and guided machine learning.
Guided machine learning
Modularizing AI brains and machine teaching are the solutions proposed by Microsoft to machine learning’s need for resource intensive training.
Modularizing the AI requirements, rather than trying to learn several complex abilities at once, reduces the learning space and allows for different approaches more suited to the problems at hand. And, by combining with guided machine learning using expert knowledge, it is possible to greatly reduce learning times.
The workshop session lasted for 60 minutes and included an extensive question and answer session. If you are already working with simulation models or considering how to take advantage of AI possibilities, we recommend watching the recording in full.
Microsoft’s Kence Anderson provides an illuminating and accessible view of how AI technologies are advancing and how they can be implemented.