reward
An operational objective signal in reinforcement learning: a numerical value associated with a state, action, transition, or trajectory and used to construct a training objective, usually through cumulative return. Optimizing with reward during training does not by itself imply that the learned system represents the reward signal as its internal goal.
Metadata
- aliases
- en
- reinforcement signal
- reward signal
- concept_id
- reward
- preferred
- en
- reward
- ja
- 報酬
- relations
- relations
- related
- status
- active
Documents: mentions (1)
prerequisite_of (6)
- goal-guarding scheming
- reinforcement learning
- reinforcement learning from human feedback
- reward hacking
- reward tampering
- training-gaming