Target Network
A technique in deep Q-learning in which the parameters of the online network are copied periodically to provide a stable bootstrapping target. It mitigates the problems of a moving target and training instability.
Core Facts
Timeline (last 90 days)
DQN uses a target network Q(s,a;θ^-) to compute the target value and periodically copies the main network's parameters to the target network, keeping the target value stable for a period of time.
Deep reinforcement learning requires stabilization techniques such as experience replay and target networks to cope with the instability and divergence risks brought by function approximation.
Value-function-based methods such as DQN indirectly derive the policy by estimating the Q value of each state-action pair, and typically need to maintain an experience replay buffer and a target network to stabilize training.
目标网络维护一份参数更新滞后的网络副本来计算TD目标值，避免训练发散
All Facts (4)
Value-function-based methods such as DQN indirectly derive the policy by estimating the Q value of each state-action pair, and typically need to maintain an experience replay buffer and a target network to stabilize training.
65%UnverifiedDQN uses a target network Q(s,a;θ^-) to compute the target value and periodically copies the main network's parameters to the target network, keeping the target value stable for a period of time.
50%UnverifiedDeep reinforcement learning requires stabilization techniques such as experience replay and target networks to cope with the instability and divergence risks brought by function approximation.
50%Unverified目标网络维护一份参数更新滞后的网络副本来计算TD目标值，避免训练发散
50%