AI 용어집
인공지능 완전 사전
Partial Observations
Scenario where demonstrations cover only a limited portion of the state space, creating unexplored areas that the agent must generalize.
Robust Policy
A learning policy designed to maintain acceptable performance when faced with partial observations and states not seen during training.
Policy Inference
Process of estimating the expert's underlying policy from a limited set of partial demonstration trajectories.
Policy Generalization
The ability of a learned policy to perform correctly in states not observed during the demonstrations, crucial for partial observations.
State Reconstruction
Technique for estimating missing or unobserved states from the partial information available in the demonstrations.
Covered State Space
The subset of the total state space actually explored in the demonstrations, defining the limits of direct imitation learning.
Learning from Demonstration
Synonym for imitation learning, specifically applied to scenarios where demonstrations are incomplete or noisy.
Out-of-Distribution Evaluation
Methodology for evaluating the policy's performance on states not present in the training data to measure its robustness.
Policy Function
Mathematical mapping π(a|s) that specifies the probability of choosing action a in state s, learned from partial demonstrations.
State Distribution
Probabilistic distribution describing the frequency of occurrence of different states in the environment, often biased in partial demonstrations.