Deep Q-learning is an extension of Q-learning that uses deep neural networks to approximate the Q-value function, particularly in environments with large or continuous state spaces. In standard Q-learning, Q-values are stored in a table, but this approach is not scalable for complex tasks. Deep Q-learning addresses this by using a deep neural network (often a convolutional neural network) to approximate Q(s, a), where the input is the state and the output is the Q-value for each action.
The deep neural network allows the agent to handle high-dimensional input spaces, such as images or sensor data, making it applicable to more complex environments like video games or robotics. The network is trained using the Q-learning update rule, and it is often coupled with techniques like experience replay and target networks to stabilize training.
Deep Q-learning has been successfully applied to a variety of challenging problems, including playing Atari games directly from pixel inputs. This method allows reinforcement learning agents to scale to environments where traditional Q-learning would struggle due to the size of the state space.