Expected generalization error = bias squared + variance + irreducible noise. Bias comes from the gap between model assumptions and the true pattern (underfitting). Variance comes from sensitivity to training data (overfitting).
As model complexity grows: bias falls, variance rises, total error is U-shaped. The sweet spot is in the middle.
Levers: reduce bias with more features / stronger model / less regularization; reduce variance with more data / regularization / ensembling. You explain it well only if you can say which term each trick moves on the U-curve.
Get asked follow-ups live, then receive a scored report — like a real MLE interview loop.
Start AI mock interview