Reinforcement Learning
Learning a policy from rewards obtained by acting in an environment.
Aperçu
Reinforcement learning trains a policy by having it act, observing the reward it receives and adjusting to get more. It solves problems where the correct action is not known in advance and only the outcome can be judged. It exists because some objectives, such as winning a game or aligning a model with human preference, are far easier to score than to demonstrate. RLHF applies it to language models.
Graphe des technologies
Comment Reinforcement Learning se rattache au reste de la pile.
Utilise et dépend de
- GRPO technique Techniques GRPO based on Reinforcement Learning High ✓ source
Utilisé par
- RLHF technique Techniques Reinforcement Learning used by RLHF High ✓ source
Appliqué à
- Chain of Thought technique Prompting Chain of Thought used for Reinforcement Learning Medium ✓ source
- Fine-tuning technique Techniques Reinforcement Learning used for Fine-tuning High ✓ source