发简信
IP属地：北京
-
Soft Actor Critic
SAC是一种off-policy方式优化统计策略的算法，来沟通统计策略优化和DDPG-style的方法。 SAC核心的特点是entropy re...
-
deep reinforcement learning：some key concepts
Action Spaces动作空间，分为离散动作空间（固定的action选项，如围棋）、连续动作空间（实值向量，如video game） Pol...
-
DDPG和TD3DDPG [Deep Deterministic Policy Gradient] Quick facts： off-policy。 只用于...
-
deep reinforcement learning: Q-learning Continuous action spaceAction a is a continous vector. 离散动作空间：如video game中选择开火、左移等离散的动作。连续动作空间：...
-
deep reinforcement learning:training a criticcirtic state value function a critic don't determine the action, given a...
-
deep reinforcement learning: Q-learningCritic critic evevaluates how good the actor is. Value Function： : usin...
-
deep reinforcement learning: DQN variantDoubel DQN Q function 倾向于高估reward的action。 Double DQN两个Q function，一个用来估计v...
-
deep reinforcement learning:on-policy off-policy PPOon-policy与环境交互的agent和学习的agent是同一个agent off-policy与环境交互的agent和学习的agent是不同...
-
李宏毅deep reinforcement learning： Policy GradientPolicy-based approach learning a actor machine learning for learning an...