Compare the Differences Between Model Building and Fine-Tuning Models CERTIFIED-AI-EXPERT Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   What is the primary difference between building a model from scratch and fine-tuning a pre-trained model?

   • ABuilding from scratch uses labeled data, while fine-tuning uses only unlabeled data.
   • BFine-tuning always produces a smaller model than building from scratch.
   • CBuilding from scratch requires a GPU, while fine-tuning can be done on a CPU only.
   • DBuilding from scratch starts with randomly initialized weights, while fine-tuning starts with weights learned on a different task.
   Reveal Solution Discussion Select an answer first
2. 2expert · hard

   A company needs a model to classify legal documents into 50 categories. They have 500,000 labeled documents. They have a limited budget and need the model to be deployed within a month. They also need the model to generalize well to new legal document formats. What is the best approach?

   • AFine-tune a small pre-trained model with aggressive regularization to prevent overfitting.
   • BUse a pre-trained model as a fixed feature extractor and train a logistic regression classifier.
   • CFine-tune a large pre-trained language model on the 500,000 documents.
   • DTrain a transformer model from scratch on the 500,000 documents.
   Reveal Solution Discussion Select an answer first
4. 3expert · hard

   A medical imaging company needs to build a model to detect a rare cancer from MRI scans. They have only 1,000 labeled scans. They have access to a large dataset of unlabeled MRI scans from other hospitals. They have a moderate compute budget and a deadline of three months. What is the best approach to maximize accuracy?

   • AUse a pre-trained model as a fixed feature extractor and train a logistic regression on the features.
   • BFine-tune a pre-trained model on the 1,000 labeled scans.
   • CPre-train a model on the unlabeled MRI scans using self-supervised learning, then fine-tune on the 1,000 labeled scans.
   • DTrain a model from scratch on the 1,000 labeled scans.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   How do data requirements typically compare between building a model from scratch and fine-tuning a pre-trained model?

   • ABuilding from scratch generally requires more data than fine-tuning to achieve comparable performance.
   • BFine-tuning requires more data than building from scratch because it must relearn all features.
   • CBuilding from scratch requires less data because it starts with random weights.
   • DBoth approaches require exactly the same amount of data.
   Reveal Solution Discussion Select an answer first
6. 5foundation · easy

   Why is fine-tuning generally less expensive than building a model from scratch?

   • AFine-tuning requires fewer training iterations and less data, reducing compute time.
   • BFine-tuning uses smaller models exclusively.
   • CFine-tuning does not require any hardware acceleration.
   • DFine-tuning is always free because it uses open-source models.
   Reveal Solution Discussion Select an answer first