A brief definition of terms used throughout this guide, phrased in such that a way that caters to a simulation-oriented audience.
Edit me
Term Definition
Reinforcement Learning (“RL”) A type of machine learning where an AI algorithm can interact with some virtual environment, by performing some actions and receiving feedback (both about the environment and the actions); it tries to learn an optimal control policy to choose optimal actions in a given circumstance.
Brain The eventual output of RL; a trained AI policy that can be queried for a (theoretically optimal) output based on a provided input.
Inkling The programming language used on the Bonsai platform for training brains. It is used to define what the AI should learn and its training regimen.
Traditional or Default model A simulation model without any RL components.
RL-ready model A simulation model that has the components necessary to be used as the training environment for an AI algorithm.
Episode A single simulation run, from start to finish; used in the context of training.
Configuration A set of variables that define a scenario or starting condition for each episode.
Observation A set of variables that describe the environment at a given point in time. This is the input to the AI algorithm.
State For the purpose of this guide, synonymous with Observation.
Action A set of variables that the AI algorithm has control over in the environment. This is the output of the AI algorithm.
Terminal [condition] A (boolean) flag or indicator for when to end an episode, allowing another to begin.
Iteration The process of taking an observation from the simulation, using it to query the AI algorithm (the brain) for an action and applying the action.
Iteration trigger An event inside the simulation model that pauses the simulation model and initiates an iteration, after which the simulation model resumes.
Bonsai Library An AnyLogic plugin/add-on library giving access to the Bonsai Connector object.
Bonsai Connector An object necessary to be added to your top-level agent for communicating with the Bonsai platform for locally hosted training.
Locally hosted training Performing RL training with the simulation model hosted on your local machine.
Uploaded training Performing RL training with the simulation model uploaded to Microsoft Azure, allowing training to be scaled.
Assessment The act of evaluating a brain’s performance across a number of episodes.
Playback A feature in the Bonsai Connector, allowing connection to an exported brain for assessment purposes.