Skip to content

Model Training & Adaptation

Policy gradient

Teaching AI to adjust its action choices toward ones expected to earn more reward.

Example

Training makes rewarded actions more likely to be chosen.

Why people use it

It teaches an agent by changing how likely it is to choose particular actions.

What you'll hear

“Make rewarded choices more likely in future attempts.”

What this means for you

Check unintended effects as well as whether the reward increases.

Can you control it?

Developer-only

The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.

Common questions

Can learning favor a shortcut people did not want?
Yes. If the reward is poorly chosen, a shortcut may score well without completing the intended task.
Does it always choose the best-looking action immediately?
No. Learning can use varied action choices to gather useful experience.
Can delayed rewards make learning harder?
Yes. It can be difficult to identify which earlier choices contributed to a later result.

Related terms

Still have questions?

Up to 500 characters.

Ask LATHIC about AI. Relevant glossary entries may be included.

Your question, the glossary entries it matches, and a rotating pseudonymous identifier go to Microsoft Azure’s OpenAI service through Vercel AI Gateway to generate an answer. Zero retention and no training are required of the provider, and LATHIC does not save your question or answer. Privacy Notice