Event

Generative Value Classifier: Q-function parameterization without action conditioning

1 reports · 1 sources

Chelsea Finn shared Alexander Swerdlow's paper, introducing a Q-function parameterization called "Generative Value Classifier." Its core feature is that the Q-function no longer requires action conditioning. The authors state that this type of method can be combined with large-scale pretrained policies and can learn and generalize better from richer output signals. The paper link is alexanderswerdlow.github.io/gvc/. The above effects are currently the authors' claims.