SESKStuff Everyone Should Know
SESK: Stuff Everyone Should Know · sesk.info/ai-high-variance-gap

Training error is far lower than test error. Which problem does that point to?

High variance, which is overfitting.

Start playing free

  1. High bias
  2. High variance ✔
  3. Irreducible noise

Why: Variance is how much the model changes with a different training set. A big gap between low training error and high test error means it over-specialized on its training data.

Huh, didn't know that: When training and test error are close but both too high, that is high bias instead.

The dad joke
My model aced the practice quiz and bombed the test. Its teacher calls that high variance; I call it cramming.
Like this card?
Play the game

References

Answer: High variance. Variance is how much the model changes with a different training set. A big gap between low training error and high test error means it over-specialized on its training data.

  1. Cornell CS4780 Lecture 12: Bias-Variance Tradeoff www.cs.cornell.edu/courses/cs4780/2018fa/lectures/lecturenote12.html
    Captures how much your classifier changes if you train on a different training set.
    Checked 2026-10-10.
  2. Cornell CS4780 Lecture 12: Bias-Variance Tradeoff www.cs.cornell.edu/courses/cs4780/2018fa/lectures/lecturenote12.html
    Training error is much lower than test error
    Checked 2026-10-10.

Think this answer is wrong? Tap "Challenge this answer" on the card in the game and tell us why.

Test yourself in the game

Artificial IntelligenceA model scores great on its training data but badly on new data. What is that called?Artificial IntelligenceWhich term measures error from a model's assumptions, even with infinite training data?Artificial IntelligenceWhat does dropout randomly remove during a training step?

That's stuff everyone should know.