Deep Q-Network

AI and Machine Learning · Reinforcement Learning · 2013 · also: DQN · deep-q-network.yaml

Replaces the Q table with a convolutional network reading pixels. The result that made reinforcement learning on raw sensory input look tractable.

• 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 Q-Learninga table cannot cover a high-dimensional state space

Referenced by

• correctsDeep Deterministic Policy Gradient fixes a defect in thisthe max over actions has no closed form once an action is a vector
• correctsDouble DQN fixes a defect in thisone network both picks the next action and rates it, so noise reads as value
• correctsDueling Network fixes a defect in thisin most states the action barely matters, yet each is estimated alone
• correctsExperience Replay fixes a defect in thisconsecutive transitions are correlated
• extendsRainbow adds capability to thisevery fix had been measured alone, against the same unimproved baseline
• correctsTarget Network fixes a defect in thisa target computed from the weights being updated moves with every step

References

• Playing Atari with Deep Reinforcement Learning — Mnih et al. — 2013 · link