Value-Based Reinforcement Learning
A reinforcement learning approach that first learns a value function, such as a Q-function, to estimate long-term return, and then extracts a policy based on the value function. Common algorithms include DQN and the critic component of SAC; it provides dense and stable learning signals in continuous control settings.
Timeline (last 90 days)
Two key factors enabling generalizable improvement are the training task distribution and the use of value-based reinforcement learning.
Value-based reinforcement learning first learns a value function to estimate the long-term return of a state or state-action pair, then extracts a policy based on that value function.
All Facts (2)
Two key factors enabling generalizable improvement are the training task distribution and the use of value-based reinforcement learning.
50%UnverifiedValue-based reinforcement learning first learns a value function to estimate the long-term return of a state or state-action pair, then extracts a policy based on that value function.
50%