Build a mind you’re proud of.
Memorising examples is not generalising
Overfitting occurs when a model fits training data but performs poorly on new data.
The idea
A model may learn accidental details rather than useful patterns. Evaluation on separate data helps expose the problem.
An example
A classifier recognises its training photos perfectly but fails on new photographs.
Put it into practice
Judge performance on examples the model did not train on.
Further reading: Google · Machine Learning ↗
Test yourself: Which result suggests overfitting?
Excellent training results and poor new-data results
The gap suggests the model learned training-specific details.
Explanation and illustrative examples by One Card Wiser. Source linked for further reading.
Open this card in the app →