Pre-Trained Models and Transfer Learning CT-AI Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   A team is building a model to classify news articles into topics. They have a small labeled dataset. They decide to use a pre-trained model like BERT and fine-tune it. Which statement best describes why BERT is a good starting point?

   • ABERT has already learned general language representations from a large corpus, which can be adapted to the news classification task.
   • BBERT can be used without any fine-tuning because it is already a classifier.
   • CBERT is a small model that requires very little data to train from scratch.
   • DBERT is specifically trained on news articles, so it already knows the topic categories.
   Reveal Solution Discussion Select an answer first
2. 2expert · hard

   A company is using a pre-trained model for a customer service chatbot. They have fine-tuned it on their own data. They are concerned about the model producing biased responses. Which combination of strategies is most effective in mitigating bias?

   • ARemove demographic features from the input data to prevent the model from discriminating.
   • BUse a larger pre-trained model to ensure it has more general knowledge.
   • CEvaluate the model's responses across demographic groups and fine-tune on a balanced, bias-audited dataset.
   • DDeploy the model without fine-tuning to preserve its original training.
   Reveal Solution Discussion Select an answer first
4. 3foundation · easy

   How does transfer learning improve performance on small datasets?

   • ABy reducing the number of parameters in the model
   • BBy using a larger model architecture that requires more data
   • CBy artificially generating more data for the target task
   • DBy providing a model that already has learned general features, which can be fine-tuned with limited data
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   Which of the following is a widely used pre-trained model primarily designed for natural language processing tasks?

   • AResNet
   • BVGG
   • CInception
   • DBERT
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A team is using a pre-trained model for a text classification task. They have a small labeled dataset. They are considering two approaches: (1) fine-tune the entire model, or (2) freeze the pre-trained layers and only train a new classifier on top. Which approach is more likely to be effective when the labeled dataset is very small?

   • AFreeze the pre-trained layers and only train a new classifier, because it reduces the risk of overfitting.
   • BBoth approaches are equally effective, so the choice does not matter.
   • CNeither approach will work because pre-trained models cannot be used for new tasks.
   • DFine-tune the entire model, because it allows the model to adapt more fully to the new task.
   Reveal Solution Discussion Select an answer first