Short Definition
Feature scaling rescales input features to comparable ranges.
Definition
Feature scaling ensures that input features contribute proportionally during training. Without scaling, features with large numeric ranges can dominate gradient updates and destabilize optimization.
Common scaling methods include normalization and standardization.
Why It Matters
Well-scaled inputs lead to faster and more stable training.
How It Works (Conceptually)
- Rescale features to fixed ranges
- Prevent dominance by large-valued inputs
- Improve gradient behavior
Minimal Python Example
scaled = (x - min_val) / (max_val - min_val)Common Pitfalls
- Scaling test data incorrectly
- Mixing scaling strategies
- Ignoring feature distributions
Related Concepts
- Normalization
- Input Preprocessing
- Training Stability
Neural Network Lexicon 