Q-Learning

Q-learning is a reinforcement learning algorithm that estimates the long-term value of taking an action in a particular state. The agent interacts with an environment, receives rewards, and updates a table or function called Q so it can favor actions expected to produce better future outcomes. It is model-free because it does not require a complete description of how the environment changes. Exploration is necessary to discover alternatives, while exploitation uses the best-known action. In large state spaces, neural networks can approximate Q values, as in deep Q-learning. Performance depends on reward design, learning rate, discounting, exploration strategy, and whether the environment remains stable enough for experience to stay relevant.