Unverified50% confidenceConceptExact time
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.
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedValue-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.76% similarUnverifiedThe network estimates Q(s,a) values for state-action pairs, using only an action-value network rather than a full policy+value dual-head structure.71% similarUnverifiedDouble DQN decouples action selection from value estimation by using the online network to select actions and the target network to evaluate Q-values, thereby alleviating the problem of Q-value overestimation.66% similarUnverifiedBackpropagation requires a global, explicit objective function to define 'what is good', and requires the computational graph of the entire network to be differentiable.65% similarUnverifiedIn standard Q Learning, the target value depends on the current Q function, causing the target value to keep changing as the parameters update, creating the "moving target" problem.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1125987API
curl https://kongchang.com/api/v1/knowledge/claims/1125987MCP
get_claim(id=1125987)