Re-enforcement learning

Jeong-rye Park is giving a talk on deep learning at the Yeungnam Math Society meeting. She it explaining how it can be used as a black box with re-enforcement learning– explaining the API, if you will.

"You set up a parameter space and an output space, and then a grading function on the output space with a reward. If the computer produces a result from the input with a high grade, you reward it."

"What's the reward?" I ask.

"You just give it a high score, we call it an R-score, R for reward."

"My computer doesn't like R-scores."

"It's just terminology."

"It likes when I clean its vents."