AI0-001 AI Concepts and Foundations Practice Question

Answer analysis

Option-by-option breakdown

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

• ✗

  Underfitting

  Why it's wrong here

  Underfitting means the model has not learned the training data, so training loss remains high; here training loss decreases steadily, indicating the network fits the training set well. It tempts because both are training-diagnostic failures, and it would be correct if both training and validation loss stayed elevated.

• ✗

  Vanishing gradients

  Why it's wrong here

  Vanishing gradients cause training loss itself to stall or fail to decrease, as early-layer weights barely update; here training loss falls steadily, so the divergence is generalisation failure. It tempts because both involve poor learning, and it would be correct if loss plateaued immediately rather than validation rising after 20 epochs.

• ✗

  Data leakage

  Why it's wrong here

  Data leakage inflates validation performance because training and validation sets share information, producing unusually low validation loss, not a steady rise while training loss falls. It tempts because leakage also causes train-validation divergence, and it would be correct if validation accuracy were suspiciously high from the outset.

• ✓

  Overfitting

  Why this is correct

  Overfitting occurs when the model memorises training data, so training loss keeps falling while validation loss rises after epoch 20. The divergence between decreasing training loss and increasing validation loss is the defining signature, indicating the network no longer generalises to unseen images.