Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Transition Model
Neural component that predicts the next latent state given the current state and action, enabling the simulation of future imaginations.
Variational Autoencoder
Neural network that encodes observations into probabilistic distribution in latent space and decodes them, enabling the generation of plausible states.
Dreamer
World Model algorithm that learns entirely in latent space using imaginations to optimize the policy without real interaction.
Imagination Rollout
Process of simulating future trajectories in latent space using the transition model to evaluate actions before execution.
State Representation
Compressed and informative encoding of raw observation capturing relevant features for decision-making.
Latent Planning
Decision process optimized in compressed latent space rather than raw observation space for computational efficiency.
Model Predictive Control
Planning strategy that sequentially optimizes actions by simulating multiple future steps through the learned model.
Curious Agent
World Model agent that intrinsically maximizes information or prediction error to efficiently explore the environment.
Augmented Imagination
Technique for augmenting training data with imagined trajectories to improve robustness and learning efficiency.
Latent Action Space
Compressed representation of actions in latent space, enabling more efficient planning and task transfer.
Hierarchical Model
Multi-scale World Model architecture capturing both fast and slow dynamics for planning at different time horizons.