Simulation modeling and learning have been around for more than a few decades. The growing integration of machine learning into end-to-end processes has resulted in more optimized operating policies and resource utilization. Skymind has partnered with AnyLogic to add reinforcement and machine learning capabilities without requiring coding skills.
Simulation has traditionally focused on optimizing one value given certain criteria. These values are used as input into policies, defined by best human logic or static rules. The promise of reinforcement learning is to determine the best logic that can be used to train the best policies given a simulation scenario and the ability to update and increase the data for retraining.
Reinforcement learning is not efficient without distributed learning. Distributed learning requires parallel machines for training deep learning algorithms and multi-agent systems for learning optimal coordination strategies. The lack of high-performance software to meet performance needs has blocked the adoption of reinforcement learning.
Successful reinforcement learning software solutions today must have the ability to take in various input types, learn in a distributed manner, provide outputs that are usable with current infrastructure, and accommodate retraining as needed with full history tracking.