Policy Training
Pathmind simplifies the process of adding AI to simulation models, from implementing an action space to orchestrating policy training in the cloud.
The Pathmind Workflow
Focus on building your simulation and Pathmind will handle the rest.
Easily Apply Reinforcement Learning to Your Simulation
Pathmind provides an easy-to-use plugin that inserts reinforcement learning tools into your simulation.
Pathmind provides a simple interface that enables you to convert simulation data into a format that a reinforcement learning algorithm can understand. No AI expertise or PhD needed.
- Defining observations needed for AI to make an informed decision.
- Setting up actions and determining when those actions are triggered, allowing AI to impact the simulation.
- Specifying metrics which embody important KPIs such as revenue and cost.
Additionally, Pathmind allows you to query the trained policy back in your simulation, enabling you to validate and showcase your results directly within your simulation.
Automated Hyperparameter Tuning and Cloud Infrastructure Orchestration
Pathmind leverages state-of-the-art automated hyperparameter tuning and distributed cloud infrastructure to achieve optimal results as quickly as possible.
- When training is executed, Pathmind will automatically provision cloud compute resources to train your policy. All the training infrastructure is handled for you.
- During training, Pathmind will dynamically adjust neural network hyperparameters on the fly to discover the best possible outcome.
- Pathmind trains your policy using the latest reinforcement learning algorithms such as PPO. We constantly test and update algorithms so you will always have access to the state-of-the-art.
No reinforcement learning knowledge or cloud infrastructure expertise is required. Pathmind handles this for you so you can focus on improving your simulation.
Manage projects and experiments with ease via the Pathmind Training interface
Execute Experiments in Parallel
Pathmind automatically orchestrates parallel experiments using the cloud.
Pathmind enables you to rapidly test dozens of reward functions in parallel. There are no restrictions on how many experiments that you can run concurrently.
- Traditionally, aspiring Data Scientists are limited to a handful of experiments at any given time, making the process of iteration slow and tedious.
- Pathmind allows parallel experimentation and automatically tracks results, enabling you to obtain a well-trained policy in days versus weeks.