reinforcement learning

A machine-learning paradigm in which an agent learns a policy for acting in an environment from evaluative feedback, typically by adjusting behavior so as to increase expected cumulative reward or another return-based objective.

Documents: mentions (2)

• 既に形成された欺瞞的な方策は、標準的な安全性訓練を経ても残り得る
• 環境に対称性があると、最適方策は選択肢を保つ方向のパワーシーキングに傾く

broader (1)

• machine learning

narrower (1)

• reinforcement learning from human feedback

prerequisite (3)

• agent
• reward
• training

prerequisite_of (1)

• reinforcement learning from human feedback