Overfitting
A model fitting its training data instead of the pattern behind it.
Overview
Overfitting is what happens when a model learns the training set well enough that its performance there stops predicting its performance anywhere else. It solves nothing; it is the default outcome of optimising a flexible model against a finite sample. It exists as the reason regularisation, held-out data and early stopping exist, and it is why a fine-tuning run on a few hundred examples can make a model worse at the task it was already good at.
Technology Graph
How Overfitting connects to the rest of the stack.
Applied to
- Dropout technique Techniques Dropout solves Overfitting High ✓ source
- Regularization concept Algorithms Regularization solves Overfitting High ✓ source
Classification
- Fine-tuning technique Techniques Overfitting part of Fine-tuning High ✓ source