AI-900 Practice Question: Describe fundamental principles of machine learning on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  The tradeoff between model accuracy and inference speed

  Why it's wrong here

  This option is incorrect because the accuracy-speed tradeoff relates to deployment considerations, such as model architecture, quantization, or hardware constraints, affecting inference latency and throughput. The bias-variance tradeoff is a purely statistical concept about prediction error sources in learning theory, independent of runtime performance. While a more complex model might be both more accurate and slower, that correlation does not define the bias-variance relationship. This option conflates engineering tradeoffs with the fundamental underfitting-overfitting dilemma.

• ✓

  The tradeoff between underfitting (high bias) and overfitting (high variance) when choosing model complexity

  Why this is correct

  This option correctly describes the bias-variance tradeoff, a foundational ML concept. As model complexity increases, bias (error from overly simplistic assumptions) decreases while variance (error from sensitivity to training data fluctuations) increases, creating a U-shaped total error curve. The optimal complexity minimizes total error by balancing underfitting (high bias, low variance) against overfitting (high variance, low bias). This tradeoff is a statistical property of learning algorithms, independent of data quantity or deployment constraints.

• ✗

  The tradeoff between training data quantity and model quality

  Why it's wrong here

  This option is incorrect because the bias-variance tradeoff concerns error decomposition at a fixed dataset size, not the relationship between data quantity and model quality. While adding more training data generally reduces variance and can improve performance, bias-variance analysis addresses how model complexity interacts with irreducible error, bias, and variance. The effect of data quantity is a separate scaling concern: with infinite data, a sufficiently flexible model can approach the Bayes error, but the complexity choice still involves bias-variance balance. Thus, this option confuses dataset management with the statistical bias-variance tradeoff.

• ✗

  The difference in fairness metrics between biased and unbiased model versions

  Why it's wrong here

  This option is incorrect because fairness metrics, such as demographic parity or equalized odds, measure bias with respect to protected attributes (e.g., race, gender) in a model's predictions. The bias-variance tradeoff, however, refers to statistical bias — systematic error in predictions due to model simplifications — and variance, the variability of predictions across training sets. Fairness bias is a societal and ethical concern about discrimination, not a property of model complexity. Therefore, comparing biased vs. unbiased model versions addresses fairness, not the statistical bias-variance tradeoff.

Go deeper

Related to this question