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.

• ✗

  How quickly the model makes predictions

  Why it's wrong here

  Accuracy has no relationship to how quickly a model produces a prediction; that speed is called inference latency and is measured in milliseconds or predictions per second. Latency is an operational efficiency concern tied to hardware and model architecture, whereas accuracy quantifies prediction correctness as the ratio of correct predictions to total predictions. Maximizing inference speed does not improve classification accuracy and can even degrade it through aggressive optimization.

• ✓

  The proportion of correct predictions out of total predictions

  Why this is correct

  Accuracy is the classification metric that calculates the proportion of correct predictions out of all predictions made, formally expressed as (true positives + true negatives) divided by the total number of evaluated cases. For instance, if a model correctly labels 90 of 100 test instances, its accuracy is 0.90, or 90%. This metric collapses both false positives and false negatives into a single correctness value, so it summarizes overall performance but must be interpreted cautiously with imbalanced classes.

• ✗

  How much memory the model uses during inference

  Why it's wrong here

  Memory usage during inference describes the model's footprint in RAM or GPU memory, influenced by parameter count, architecture, batch size, and optimization techniques like quantization from float32 to float16 or int8. That resource-consumption metric affects deployment choices such as edge-device feasibility or cloud cost, but it is entirely distinct from accuracy, which measures how often predicted labels match the ground-truth labels. A model can be highly memory-efficient yet have poor accuracy, because memory footprint and predictive correctness answer different questions.

• ✗

  The number of training examples used to build the model

  Why it's wrong here

  The number of training examples used to build the model reflects the training dataset size, which can affect learning quality, overfitting, and generalization, but it is not the definition of accuracy. Accuracy is evaluated on a labeled test or validation set as the percentage of correct predictions out of all predictions made, not on the training count itself. Even a large training corpus can produce low accuracy if the model fails to generalize, so accuracy remains an independent measure of prediction performance.

Go deeper

Related to this question