Compare the Differences Between Model Building and Fine-Tuning Models CERTIFIED-AI-EXPERT Practice Questions (Page 3)

Questions 11–15

1. 11foundation · easy

   Which statement best compares the computational resources needed for building a model from scratch versus fine-tuning?

   • AFine-tuning typically requires more computational power because it must train on multiple datasets.
   • BBoth approaches require identical computational resources.
   • CBuilding from scratch requires less time because it starts with random weights.
   • DBuilding from scratch typically requires more computational power and time than fine-tuning.
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   How does fine-tuning typically affect a model's generalization compared to building from scratch when the target dataset is small?

   • AFine-tuning always leads to overfitting on the target dataset.
   • BFine-tuning usually worsens generalization because it forgets the original task.
   • CFine-tuning usually improves generalization because the model starts with general features.
   • DFine-tuning has no effect on generalization.
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   Which statement accurately describes fine-tuning a pre-trained model?

   • AIt involves training a model exclusively on unlabeled data to learn general representations.
   • BIt involves only changing the model's architecture without any additional training.
   • CIt involves discarding all pre-trained weights and retraining the model from scratch on new data.
   • DIt involves taking a model already trained on a general task and further training it on a specific, related task.
   Reveal Solution Discussion Select an answer first
5. 14foundation · easy

   What is a common performance trade-off when fine-tuning a pre-trained model instead of building from scratch?

   • AFine-tuning produces models that are less accurate but faster at inference.
   • BFine-tuning always generalizes worse than building from scratch.
   • CFine-tuning always achieves higher accuracy than building from scratch.
   • DFine-tuning may perform poorly if the pre-trained model's original task is very different from the target task.
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   A company is building a speech recognition model for a very specific dialect that has no existing pre-trained model. They have a large dataset of 5 million hours of audio in that dialect. They have a substantial compute budget. What is the most appropriate approach?

   • AUse a pre-trained model as a fixed feature extractor and train a language model on the extracted features.
   • BFine-tune a small pre-trained model with a low learning rate for a few epochs.
   • CTrain a speech recognition model from scratch on the 5 million hours.
   • DFine-tune a general-purpose speech recognition model on the 5 million hours.
   Reveal Solution Discussion Select an answer first