single mention · post-training
Hierarchical penalty escalation
Escalates penalties or rejects data at progressively larger levels when no usable turns, contexts, or sequences remain.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See reinforcement learning algorithm for the methods filed beside it.
Also called Penalties escalate along the hierarchy.
Filed under post-training › reinforcement learning algorithm