Techniques & Methods
Offline Reinforcement Learning
Offline RL (also called batch RL) trains agents entirely on pre-collected datasets, making it valuable when live environment interaction is costly or dangerous—such as healthcare, autonomous driving, or robotics. The agent must learn good policies from static data without the ability to explore.
Key challenges include distributional shift (the offline data may not cover the situations the learned policy encounters) and the overestimation of Q-values. Conservative offline RL methods address these by being pessimistic about out-of-distribution actions.
Authority Links
Related Terms
Techniques & Methods
Reinforcement Learning
An agent learns by taking actions in an environment and receiving rewards or penalties.
Techniques & Methods
Reinforcement Learning from Human Feedback (RLHF)
Training technique that refines AI models using feedback from human evaluators on output quality.
Techniques & Methods
Markov Decision Process
Mathematical framework modeling sequential decision-making in environments with probabilistic outcomes.
Miscellaneous
Training Data
The labeled or unlabeled dataset used to fit a model's parameters during the learning process.