Dueling Network

AI and Machine Learning · Reinforcement Learning · 2016 · dueling-network.yaml

Splits the head into a state-value stream and a per-action advantage stream, then adds them back. How good a state is gets learned once instead of separately inside every action's estimate.

• supersedescorrects · extends
• classifiesspecializes · part-of
• substitutes forapproximates · alternative-to
• depends onrequires · validates

Colour is the family; a dashed line is the second member of it.

This node

• correctsfixes a defect in Deep Q-Networkin most states the action barely matters, yet each is estimated alone

References

• Dueling Network Architectures for Deep Reinforcement Learning — Wang, Schaul, Hessel, van Hasselt, Lanctot, de Freitas — 2016 · link