Pre-Trained Models and Transfer Learning CT-AI Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A team is using a pre-trained language model to build a resume screening tool. They notice that the model consistently ranks candidates from certain demographic groups lower. What is the most likely cause of this issue?

   • AThe model is outdated and does not recognize modern resume formats.
   • BThe model is too large and overfits the training data, causing random performance differences.
   • CThe model was trained on biased data, and the bias is embedded in the pre-trained representations.
   • DThe model is not fine-tuned on enough domain-specific data, so it cannot understand resumes at all.
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A bank is using a pre-trained language model to assess loan applications. The model was trained on historical data that reflects past lending decisions, which were biased against certain ethnic groups. The bank is required to ensure fair lending practices. Which approach is most effective in mitigating the bias risk?

   • ARemove demographic features from the input data to ensure the model cannot discriminate.
   • BDeploy the model as-is and rely on human review for all decisions.
   • CUse a larger pre-trained model to reduce bias because it has more general knowledge.
   • DFine-tune the model on a balanced, bias-audited dataset and evaluate its decisions across demographic groups.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A data science team fine-tunes a pre-trained model on a small dataset of legal documents. After fine-tuning, the model performs well on the legal documents but its performance on general language understanding tasks drops significantly. What is this phenomenon called?

   • ADomain shift
   • BOverfitting
   • CCatastrophic forgetting
   • DNegative transfer
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A startup is building a sentiment analysis tool for customer reviews in the hospitality domain. They have only 2,000 labeled reviews and limited GPU budget. They decide to start with a pre-trained BERT model and fine-tune it on their data. Which statement best describes the primary benefit of this approach compared to training a model from scratch?

   • AIt allows the model to be deployed without any further training or validation.
   • BIt guarantees higher accuracy than any model trained from scratch on the same data.
   • CIt eliminates the need for any domain-specific fine-tuning because BERT already understands all domains.
   • DIt reduces the amount of labeled data and computational resources needed to achieve good performance.
   Reveal Solution Discussion Select an answer first
6. 10foundation · easy

   What is transfer learning in the context of machine learning?

   • AA technique where knowledge gained from solving one problem is applied to a different but related problem
   • BA method for transferring data between different storage systems
   • CA process of moving a trained model from one server to another
   • DA technique for converting a model from one programming language to another
   Reveal Solution Discussion Select an answer first