Technical Note: Q-Learning

By Christopher J. C. H. Watkins, Peter S. Dayan. 2004

2,385 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Reinforcement Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Reinforcement Learning

Citations in the corpus

Deep Reinforcement Learning for Autonomous Driving: A Survey (2020, background): “The theoretical guarantees of Q-learning hold with any arbitrary initial Q values [21]; therefore the optimal Q values for a MDP can be learned by starting with any initial action value function estimate.”
Factorized Q-learning for large-scale multi-agent systems (2018, methodology): “For example, Nash-Q [5] and Friend-or-Foe [9] extend the classic Q-learning [30] by encoding the interactions between environment and agents in a so-called Nash Q function.”
Guided Deep Reinforcement Learning for Swarm Systems (2017, methodology): “Q-learning [17] is applied to learn the policy.”
Independent reinforcement learners in cooperative Markov games: a survey regarding coordination problems (2012, methodology): “Especially, several modiﬁcations to the Q-learning algorithm (Watkins & Dayan, 1992) have been proposed for multi-agent environments.”
State of the Art Control of Atari Games Using Shallow Reinforcement Learning (2015, methodology): “DQN is based upon the well-known Q-learning algorithm [31] and uses a CNN to”
A Multiple-Goal Reinforcement Learning Method for Complex Vehicle Overtaking Maneuvers (2011, methodology): “QL by Watkins [26] is a simple model-free RL approach that allows the agent to learn to optimally act in the Markovian domain.”
Generalization and Regularization in DQN (2018, background): “Q-learning (Watkins and Dayan, 1992) is a traditional approach to learning an optimal policy from samples obtained from interactions with the environment.”
Value-Difference Based Exploration: Adaptive Control between Epsilon-Greedy and Softmax (2011, background): “However, if the stochasticity in action selections becomes zero (i.e. greedy), Sarsa technically becomes the same as Q -learning, and thus also convergent under several conditions [13].”