AI-ordlista
Den kompletta ordlistan över AI
Meta-Reinforcement Learning
Reinforcement learning approach where the agent learns to learn, acquiring meta-knowledge to quickly adapt to new tasks with few experiences.
Meta-Learner
Algorithm or model that optimizes a learning process to acquire rapid adaptation capabilities to new tasks not seen during training.
Task-Specific Policy
Reinforcement learning policy adapted to a particular task, quickly generated by the meta-learner from few experiences.
Proximal Meta-Policy Optimization (ProMP)
Meta-RL algorithm that extends PPO to meta-learning, optimizing a meta-policy capable of generating task-specific policies.
Meta-World
Benchmark and standardized environment to evaluate meta-RL algorithms on robotic manipulation tasks with varied task distribution.
RL² (Reinforcement Learning Squared)
Meta-RL framework where the reinforcement learning algorithm itself is learned by another RL process, integrating history into the agent's state.
Meta-Experience Replay
Experience buffer technique organized by tasks to facilitate rapid adaptation and knowledge transfer between different tasks.
Meta-Policy Gradient
Optimization algorithm that calculates gradients with respect to meta-parameters to improve expected performance on the task distribution.
Hindsight Experience Replay (HER) in Meta-RL
Extension of HER to meta-RL where experiences are reinterpreted with different objectives to improve sampling and inter-task generalization.
Curriculum Learning in Meta-RL
Progressive sequencing of training tasks by increasing complexity to improve the adaptation capability of the meta-learner.
Meta-Imitation Learning
Combination of meta-learning and imitation learning where the agent learns to quickly imitate new demonstrations with few examples.
Meta-Off-Policy Evaluation
Evaluation of the performance of a meta-learned policy on new tasks using only previously collected off-policy data.