PRODUCTION SCENARIO
An energy utility is tuning a building's cooling setpoints in simulation. There is no record of correct settings: the controller earns credit for every hour it holds temperature in range at lower power draw, loses credit when the room drifts out of range, and improves over thousands of simulated days.
Which machine learning approach describes this training?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Reinforcement learning
Google describes reinforcement learning as learning by doing through trial and error: an agent attempts a task, receives positive reinforcement when it performs well and negative reinforcement when it performs poorly, and improves over many attempts. That is the credit-and-penalty loop here. Supervised learning is the closest miss, because the credit score looks like feedback about an answer - but a label states what the correct setpoint was, while a reward only says how well the attempt turned out. Unsupervised learning finds structure in a dataset it is handed, and there is no dataset here, only the consequences of acting. Deep learning is not a fourth approach alongside these; it describes the network architecture, which a reinforcement learning agent may or may not use.
Official Source: https://cloud.google.com/learn/what-is-machine-learning ↗