Скачать или смотреть 🎧 Reinforcement Learning for AI

Reinforcement learning is an approach that trains artificial intelligence to make decisions based on evaluative feedback and rewards. It can shape decision making in the ai systems we use in our daily lives including chatbots, autonomous vehicles and even your T.V. streaming recommendations. We’ll explore how reinforcement learning is helping train humans in skills we usually spend years perfecting, in the U.S. National Science Foundation’s “Discovery Files”.

“The key idea here is to learn through trial and error guided by some kind of reward. So after each actions have been made what are the incentives or penalties that are received. And then the policy is what strategy that the agents eventually learn.”

Mingyi Hong is a professor at the University of Minnesota where his NSF-supported lab works to optimize artificial intelligence systems. One strategy, called inverse reinforcement learning, flips the script by training AI directly from human expert demonstrations.

“So instead of learning the policy from the reward, learn the reward function itself by observing expert behavior.”

This teaches AI to model expert human actions—like how a skilled surgeon operates, adjusts or reacts in real time. The goal is to develop AI systems that can potentially train the next generation of American workers, especially in fields like health care and advanced manufacturing.

Learn more about this research by listening to "NSF's Discovery Files" podcast, Training Artificial Intelligence Experts.