midQ-Learning
What is function approximation in Q-Learning?
Updated May 17, 2026
Short answer
Function approximation replaces Q-tables with models like neural networks to handle large state spaces.
Deep explanation
Instead of storing Q-values in a table, a function approximator (like a neural network) estimates Q(s,a). This enables handling continuous or very large state spaces where tabular methods fail.
Real-world example
Used in Atari game agents where pixel inputs are too large for Q-tables.
Common mistakes
- Assuming tabular Q-learning scales to high-dimensional inputs.
Follow-up questions
- Why use neural networks in Q-learning?
- What is overfitting in RL?