This post has been republished via RSS; it originally appeared at: New blog articles in Microsoft Community Hub.
Project Bonsai
- Improved action masking. Action masking helps to guarantee valid behavior and speed up training by restricting the available AI actions based on the current state of the system. For example, it is now easier to train a system that knows not to route work to a piece of equipment that is down for maintenance. (Learn more)
- [BETA] Train AI with data. You can now train policies directly from data without the need for detailed simulation. Training from data works best when you have a training dataset made up of mostly expert-level decisions. For example, a dataset curated by experienced operators running a process. Rather than copying the behavior described in the training dataset, the brain learns how to optimize operations by combining learnings across all decisions recorded in the data. If you want to improve on a control problem for which there is a reasonable existing solution, or if some of your operators seem to outperform others and you cannot figure out how they do so, then you should consider training with data. (Learn more)
- Model Based Training Accelerator (MBTA). Accelerate brain training with slow simulation! When you enable MBTA, Bonsai learns the dynamics of your simulation using a neural-network model and uses the model to drive brain training in conjunction with the original simulation. Speed improvements depend on the speed of your original simulation is, but early use has shown a training time reduction of 50%! Assessments are always done with the original simulation and the MBTA feature is turned on/off dynamically to ensure high quality brains. (Learn more)
- Assessment 2.0 offers richer AI analysis tools. New assessment tools help users evaluate their AI model performs against training goals with improved performance insights on model performance. Assessment 2.0 tools are available during and after training completes. Use Assessment 2.0 tools to review metrics and visualizations that help you understand model behavior and export your assessment data with a few clicks to perform advanced/custom analyses in an environment of your choice. (Learn more)
- Bonsai samples in the Azure marketplace. Bonsai samples are designed to be a rich starting point to help you get started with Machine Teaching. Samples provide example Inkling, preconfigured simulation models, walkthroughs, and (in some cases) video demonstrations of the sample in action. Install samples in your Bonsai workspace to see brain training in action and learn about what is possible with the Microsoft Bonsai platform. We have 5 samples published already with more on the way: