Using a technique we call Machine Teaching, leverage your subject matter expertise to deconstruct complex problems into the key concepts you want an AI model to learn. These concepts are programmed into a model using Bonsai’s special-purpose programming language, Inkling.
Your Inkling code is then combined with a simulation of your real-world system and fed into the Bonsai AI Engine for training.
Based on the Inkling program, the AI Engine automatically selects the best deep reinforcement learning algorithm to train an AI model, or BRAIN, in simulation. No laying out neural nets or tuning hyperparameters required.
A trained BRAIN can then stream predictions to your application, drastically improving the operations of your real-world system. Each BRAIN can be continually managed, debugged, and improved upon from within the Bonsai Platform.
Machine Teaching is the fundamental abstraction needed to combine your subject matter expertise with machine learning algorithms, enabling more efficient and accurate model development. Using Machine Teaching, complex problems are broken down into simpler concepts that are trained individually before being combined to solve the end objective. This approach significantly decreases model training time and allows for reusability of each individual concept.
Simulations and digital twins are replicas of real-world systems. These tools provide an environment to train the AI, aka a Bonsai BRAIN, to solve specific problems using deep reinforcement learning. Bonsai’s integrated tooling allows enterprises to quickly connect any existing or custom simulator into the platform for training; including MATLAB Simulink (engineering, manufacturing), Transys (energy), Gazebo (robotics) and AnyLogic (supply chain).
Each Machine Teaching program, along with the specified simulation, is fed into the Bonsai AI Engine, where it is compiled to automatically generate and train the best machine learning model for a given problem. The resulting high-level model, aka BRAIN, can then be connected into your hardware or software application through Bonsai provided libraries.
Bonsai BRAINs are trained AI models with the ability to intelligently control or optimize real-world systems. BRAINs, built with cutting-edge deep reinforcement learning algorithms and trained in simulation, can be deployed on-premises, in the cloud or at the edge. Each BRAIN is available for ongoing debugging and refinement, and can be repurposed for use in other applications.