Fundamentals of Generative AI →hardMultiple Choice

Generative AI Leader Fundamentals of Generative AI Practice Question

Exhibit

Refer to the exhibit.
```
INFO: Training started.
INFO: Epoch 1/10 - loss: 2.3456 - accuracy: 0.1234
INFO: Epoch 2/10 - loss: 1.9876 - accuracy: 0.2345
INFO: Epoch 3/10 - loss: 1.6543 - accuracy: 0.3456
INFO: Epoch 4/10 - loss: 1.4321 - accuracy: 0.4567
INFO: Epoch 5/10 - loss: 1.2345 - accuracy: 0.5678
INFO: Epoch 6/10 - loss: 1.0987 - accuracy: 0.6789
INFO: Epoch 7/10 - loss: 0.9876 - accuracy: 0.7890
INFO: Epoch 8/10 - loss: 0.8765 - accuracy: 0.8901
INFO: Epoch 9/10 - loss: 0.7654 - accuracy: 0.9012
INFO: Epoch 10/10 - loss: 0.6543 - accuracy: 0.9123
```

Refer to the exhibit. A data scientist is fine-tuning a model. The training loss and accuracy are improving each epoch. However, after training, the model performs poorly on a held-out validation set. What is the most likely issue?

⚠ Common exam trap

Google Cloud often tests the distinction between overfitting and underfitting by presenting improving training metrics alongside poor validation performance, which candidates may misinterpret as a learning rate issue or data leakage if they do not recognize the hallmark divergence pattern.

Answer choices

Why each option matters

Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.

Correct answer & explanation

✓

Overfitting

The model's training loss and accuracy improve each epoch, but performance on the validation set is poor. This classic symptom indicates overfitting, where the model memorizes the training data (including noise) rather than learning generalizable patterns. In fine-tuning, this often occurs when the model is trained for too many epochs or the dataset is too small relative to model capacity.

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 both training and validation performance remain poor, yet the exhibit shows training loss and accuracy still improving each epoch, contradicting that diagnosis. It is tempting because poor validation results superficially resemble insufficient learning, and underfitting is the correct answer when training metrics themselves plateau at a low level.
✗
Inappropriate learning rate
Why it's wrong here
An inappropriate learning rate would disturb the training curves themselves, causing oscillation, divergence, or stalled loss, but the exhibit shows training metrics improving steadily. It is tempting because learning rate is the usual suspect whenever training behaves unexpectedly, and it is correct when loss fails to descend or diverges rather than when generalisation gaps appear.
✗
Data leakage
Why it's wrong here
Data leakage contaminates training with information from the validation distribution, which typically inflates validation scores, not depresses them; here validation performance is poor while training improves. It is tempting because leakage is a common cause of misleading evaluation results, and it is correct when validation metrics look implausibly strong relative to deployment.
✓
Overfitting
Why this is correct
Overfitting occurs when the model memorises training data, so training loss and accuracy keep improving while generalisation to unseen data degrades. That mismatch between strong training metrics and poor held-out validation performance is precisely the stem's symptom.

Go deeper

Generative AI Leader Fundamentals of Generative AI Practice Question

Exhibit

Why each option matters

Option-by-option breakdown

Related to this question