跳到主要内容
Learn Motion
探索课程产品能力价格

下载 Learn Motion

用手机扫描二维码,下载 App。

App Store

扫码从 App Store 下载 Learn Motion打开商店页面

Google Play

扫码从 Google Play 下载 Learn Motion打开商店页面
登录
Learn Motion

How AI Learns from Rewards and Mistakes

1The Trial-and-Error Loop: How an Agent Learns2Formalizing the Problem: States, Actions, Rewards, and Policies3Measuring Success: Return, Discounting, and Value4Learning from Experience: Value Updates and Q-Learning5Exploring vs. Exploiting: Where Mistakes Come From6Learning Policies Directly and Scaling Up
The Trial-and-Error Loop: How an Agent Learns

One Turn of the Loop

1 / 2
Watch the cycle move. The agent reads the observation, picks an action, and the environment responds with a reward and a new observation. Notice where the loop closes: the reward feeds the update, and the update changes the next action. That update is the only place learning happens. If you removed it, the agent would keep behaving identically no matter how many rewards it collected.
0:00 / 0:00

An agent and an environment exchange information in a repeating cycle. The agent observes the current situation, chooses an action, and sends that action to the environment. The environment changes and returns a reward together with the next observation. The agent then uses that reward to adjust how it will choose actions in the future, and the cycle begins again from the new observation.

Each element has a distinct job. The observation is the agent's only window on the world; whatever it cannot observe, it cannot use. The action is the agent's only way to influence what happens next. The reward is a single number scoring the immediate consequence of the action just taken. The update is where learning actually occurs: without it, the agent would repeat the same behavior forever and nothing would improve.

The learning signal enters at the reward and is consumed by the update. A reward arriving after an action tells the agent that the outcome was good or bad, but it arrives attached to the whole situation, not to a specific decision. That ambiguity is the central difficulty of reinforcement learning and the reason the update step must reason about credit rather than simply copy an answer.

参考资料

  1. [1]Reinforcement learning (Wikipedia)en.wikipedia.org
  2. [2]Reinforcement Learning: An Introduction (Sutton & Barto, 2nd ed.)incompleteideas.net
1 / 2下一页

Learn Motion

生成课程,认真学会。

运营方:Wuhan Daoyin Technology Co., Ltd.

联系我们: [email protected]
隐私政策用户服务协议

© 2026 Learn Motion