Pre-Trained Models and Transfer Learning CT-AI Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A team fine-tunes a pre-trained model on a very small dataset of medical notes. After fine-tuning, the model performs well on the training data but poorly on new medical notes. What is the most likely risk that occurred?

   • ACatastrophic forgetting
   • BNegative transfer
   • CBias amplification
   • DOverfitting
   Reveal Solution Discussion Select an answer first
2. 12application · medium

   A company wants to use a pre-trained model for a customer service chatbot. They are concerned about the model producing biased or inappropriate responses. Which mitigation strategy is most directly aimed at reducing this risk?

   • AFine-tuning the model on a curated dataset of domain-specific, bias-audited conversations.
   • BDeploying the model without any fine-tuning to preserve its original training.
   • CUsing a larger pre-trained model to ensure it has more general knowledge.
   • DIncreasing the learning rate during fine-tuning to speed up adaptation.
   Reveal Solution Discussion Select an answer first
4. 13application · medium

   A company is using a pre-trained model to analyze social media posts for brand sentiment. The model was trained on data from 2019. They notice that the model misinterprets recent slang and references. What is the most likely limitation of the pre-trained model causing this issue?

   • ACatastrophic forgetting
   • BBias in the training data
   • COutdated knowledge
   • DLack of domain specificity
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   A company is fine-tuning a pre-trained model for a niche domain. They have a small labeled dataset. They are concerned about both overfitting and catastrophic forgetting. Which approach best balances these risks?

   • AFreeze all layers and only train a new classification head.
   • BFine-tune with a high learning rate to quickly adapt to the niche domain.
   • CUse a lower learning rate and include a small portion of general data during fine-tuning.
   • DTrain the model from scratch on the small dataset to avoid transfer learning risks.
   Reveal Solution Discussion Select an answer first
6. 15foundation · easy

   Which of the following is a primary benefit of using transfer learning?

   • AReduced training time and lower data requirements for the target task
   • BGuarantees better performance than training from scratch in all cases
   • CRemoves the need for model evaluation on the target task
   • DEliminates the need for any labeled data in the target task
   Reveal Solution Discussion Select an answer first