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.

• ✗

  To provide the primary examples for training the model's weights

  Why it's wrong here

  The validation dataset does not provide the primary examples for updating the model's weights. During supervised training, the optimizer adjusts weight values by minimizing a loss function computed from the training dataset, not from the validation split. Validation examples are only passed through the model in evaluation mode to gauge current performance, so they never directly contribute to gradient updates. Therefore, describing validation as the source of training examples conflates it with the actual training set.

• ✓

  To tune hyperparameters and monitor performance during training without using test data

  Why this is correct

  The validation dataset exists to provide ongoing feedback during active model development, letting you tune hyperparameters, compare training runs, and detect overfitting or underfitting as training proceeds. Because it is a separate labeled split that is not used for gradient computation, its loss and accuracy offer an unbiased-in-the-loop signal that reflects how well the model generalizes beyond the training set. Crucially, using a validation set keeps the test dataset completely untouched for the final evaluation, preventing leakage from tuning choices. This makes validation an essential part of the training workflow rather than a storehouse, trainer, or final grader.

• ✗

  To provide the final, unbiased assessment of model performance

  Why it's wrong here

  The final, unbiased assessment of model performance must come from an untouched test subset, not validation data. Because validation is used iteratively throughout development to tune hyperparameters and select among checkpoints, the model and developer can implicitly adapt to validation-set patterns, making it a biased estimate of real-world performance. The test dataset remains completely hidden until the end precisely to give a one-time, honest measurement. So validation is not the right tool for the final unbiased evaluation.

• ✗

  To store the model's trained weights for later use

  Why it's wrong here

  The validation dataset is in no way responsible for storing the model's trained weights. Once training completes, learned weights are serialized into model artifact files (for example, .pkl or .h5 formats) for later deployment or resumption. During development, validation data is a labeled subsample used to compute metrics after epochs or runs, and it exists independently of the weights that are saved. Hence, this option confuses the model persistence mechanism with the evaluation role of a validation set.

Go deeper

Related to this question