What is the difference between underfitting and overfitting in Deep Learning?
Updated May 16, 2026
Short answer
Underfitting occurs when a model cannot learn training patterns, while overfitting occurs when it memorizes training data and fails to generalize.
Deep explanation
Underfitting and overfitting represent opposite extremes of model learning.
Underfitting:
- Model is too simple.
- High training and validation error.
- Fails to capture underlying relationships.
- Causes include insufficient layers, weak features, or inadequate training.
Overfitting:
- Model is too complex.
- Very low training error but high validation error.
- Learns noise instead of general patterns.
- Common in deep networks with excessive parameters.
The goal is balanced generalization where both training and validation performance are strong.
Solutions for underfitting:
- Increase model complexity.
- Train longer.
- Add features.
Solutions for overfitting:
- Dropout.
- Regularization.
- Data augmentation.
- Early stopping.
Real-world example
An image classifier that memorizes training images but fails on real-world camera images is overfitting.
Common mistakes
- Judging model quality only using training accuracy instead of validation metrics.
Follow-up questions
- Why are deep networks prone to overfitting?
- What is model generalization?
- How does early stopping prevent overfitting?