Learn why strong training performance can still hide a weak model, and how train, validation, and test behavior reveal what is really happening.
Listen to hear this room section by section.
Key Ideas
Work through these sections in order. Each one builds the mental model you need before the checkpoint questions will feel easy.
Underfitting usually means the model or training setup has not captured enough signal to perform well even on the training data. Overfitting is different: the model fits the training data too specifically and fails to carry that performance to fresh examples.
In practice, these regimes often show up as different train-versus-validation patterns. That comparison is more informative than either score on its own.
Check Your Understanding
These checkpoints reinforce the lesson you just read. If one feels fuzzy, reopen the relevant section above before trying again.
Task 1
Choose the situation that best represents overfitting.
Task 2
Match each score pattern to the best diagnosis.
Task 3
Write a short explanation of how a validation set helps builders make better choices.
Ready To Move On?
Up next: Evaluation Debug Lab