Bias-Variance Tradeoff Quiz

Answers and explanations appear as you take the quiz.

1. A model that is too simple to capture the underlying pattern in the data primarily suffers from:
   • High variance
   • High bias
   • Low irreducible error
   • Overfitting
2. Which scenario is a classic sign of high variance (overfitting)?
   • Training and test error are both high and similar
   • The model ignores most features
   • Adding more regularization increases training error toward test error
   • Training error is very low but test error is much higher
3. In the classic bias-variance picture described in the guide, which tradeoff can accompany increased model flexibility?
   • Bias decreases and variance increases
   • Both increase
   • Both decrease
   • Bias increases and variance decreases
4. Adding L2 (ridge) regularization to a model generally:
   • Reduces bias at the cost of more variance
   • Eliminates irreducible error
   • Reduces variance at the cost of slightly more bias
   • Has no effect on either bias or variance
5. The irreducible error in the bias-variance decomposition represents:
   • Error caused by choosing the wrong algorithm
   • Error from inherent noise in the data that no model can remove
   • Error that disappears with more training data
   • Error caused only by high variance
6. The 'double descent' phenomenon observed in deep learning suggests that:
   • Test error can decrease again as models grow far past the point of interpolating the training data
   • Test error always follows a simple U-shape
   • Variance never matters for neural networks
   • Bias is always zero in deep models
7. In a squared-error decomposition, squared bias is 2, variance is 3, and irreducible noise is 1. What is the total expected error?
   • 14
   • 5
   • 3
   • 6
8. Why can averaging many de-correlated trees help in the random-forest example?
   • It removes all noise from the data-generating process
   • It guarantees zero test error
   • It can reduce the variance of the combined prediction
   • It converts every tree into a linear model