2024 · Artificial intelligence
Reinforcement learning connects actions, rewards and improving predictions.
Visual guide
Choose an action in an environment.
Receive a reward and a new state.
Revise predictions and future choices.
A reward signal specifies a training objective; maximising it does not automatically produce the behaviour a person intended.
01 · The idea
Andrew Barto and Richard Sutton established conceptual and algorithmic foundations of reinforcement learning. An agent acts, observes what happens and uses reward signals to improve future behaviour. Temporal-difference learning updates a prediction using both an immediate outcome and a revised prediction of what follows.
Learning through interaction became a central approach to decision-making systems, from games to control.
02 · The people
Andrew Barto helped establish reinforcement learning as a computational approach to learning from rewards. At the University of Massachusetts Amherst, his work with Richard Sutton and others connected adaptive control, neural networks, and learning through interaction. He contributed to the conceptual and algorithmic foundations of the field and co-authored its standard textbook. His role in the joint award reflects both foundational research and decades of mentorship and synthesis.
Read the biography ↗
Richard Sutton developed central ideas in reinforcement learning, including temporal-difference methods that learn predictions from the difference between successive estimates. His work also investigated how learning, planning, and action can be integrated within one agent. He collaborated with Andrew Barto on foundational research and their widely used textbook. His affiliations at Alberta and Keen Technologies reflect continued work on general learning systems after the field became widely influential.
Read the biography ↗03 · Read further
Original explanations based on the official award record. The subject label is an editorial navigation aid.
Deep networks learn useful representations through many layers of computation.
The story
Probabilistic and causal models made different kinds of questions explicit.
The story