|
|
|
Reinforcement learning involves three important concepts: the agent, some states, and a set of actions per state. By executing an action in a specified state, the agent is scored with a reward. Again imagine the computer game Super Mario. You are Mario, you are in a game level, standing next to a cliff edge. Above you is a coin. You being Mario, in a game level, at a specific position ... that's your state. Moving one step to the right (an action) will take you over the edge, that would give you a low numerical score. However, pressing the jump button, you would score a point and you would be alive. That's a positive outcome and that should award you a positive numerical score.
|
|
|
|
Reinforcement learning involves three important concepts: the agent, some states, and a set of actions per state. By executing an action in a specified state, the agent is scored with a reward. Again imagine the computer game Super Mario. You are Mario, you are in a game level, standing next to a cliff edge. Above you is a coin. You being Mario, in a game level, at a specific position ... that's your state. Moving one step to the right (an action) will take you over the edge, that would give you a low numerical score. However, pressing the jump button, you would score a point and you would be alive. That's a positive outcome and that should award you a positive numerical score.
|
|
|
|
The point of all this is that by using reinforcement learning and a simulator (the game), you can learn how to play the game to maximize the reward which is staying alive and scoring as many points as possible.
|
|
|
|
The point of all this is that by using reinforcement learning and a simulator (the game), you can learn how to play the game to maximize the reward which is staying alive and scoring as many points as possible.
|