Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Differentiable Model-Based RL
Approach of reinforcement learning where gradients are backpropagated through a differentiable environment model to directly optimize policies.
Neural Network Dynamics Models
Neural networks trained to predict the evolution of environment states based on actions, enabling differentiable simulation.
Differentiable Optimization
Technique where optimization layers are differentiable, enabling end-to-end learning of nested optimization processes.
Imagined Experience Replay
Generation of synthetic trajectories through a differentiable model to enrich the learning experience without real interaction.
Differentiable Planning
Planning algorithms where each step is differentiable, enabling continuous optimization of plans through gradient descent.
Differentiable Simulator
Simulation environment implemented with differentiable operations, enabling gradient calculation throughout the entire simulation.
Policy Gradient Through Model
Method that calculates policy gradients by propagating rewards through a differentiable environment model.
Differentiable Environment Models
Environment models specifically designed to support gradient backpropagation through their internal operations.
Differentiable Game Engines
Game engines modified to support differentiation, enabling reinforcement learning in complex environments.
Neural ODEs in RL
Use of neural ordinary differential equations to model the continuous dynamics of the environment in a differentiable way.
Differentiable Control Theory
Application of theoretical control principles implemented in a differentiable way for end-to-end learning of controllers.