Pre-Trained Models and Transfer Learning CT-AI Practice Questions (Page 4)

Questions 16–20

1. 16foundation · easy

   In which application domain is the pre-trained model ResNet most commonly used?

   • ANatural language processing
   • BSpeech recognition
   • CTime series forecasting
   • DImage recognition and computer vision
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   How can fine-tuning on domain-specific data help mitigate risks of using a pre-trained model?

   • ABy completely removing all biases from the model
   • BBy increasing the model's training time to ensure it learns everything
   • CBy adapting the model's knowledge to the specific patterns and terminology of the target domain
   • DBy making the model larger so it can store more domain knowledge
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A medical imaging team wants to build a model to detect lung nodules in chest X-rays. They have a small dataset of annotated X-rays. Which pre-trained model would be the most appropriate starting point?

   • AGPT-3, because it is a large language model that can generate text descriptions of images.
   • BBERT, because it is a transformer model that can process both text and images.
   • CWord2Vec, because it provides vector representations of words that can be adapted to images.
   • DResNet, because it is a convolutional neural network pre-trained on ImageNet and can be fine-tuned for image classification.
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   What is 'negative transfer' in the context of transfer learning?

   • AWhen applying knowledge from the source task actually worsens performance on the target task
   • BWhen the model forgets the source task after fine-tuning
   • CWhen the target task has more data than the source task
   • DWhen the pre-trained model is too large to be deployed
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   What is a potential risk of using a pre-trained model that was trained on a dataset with inherent biases?

   • AThe model will only work on the original training data
   • BThe model will become completely unusable for any task
   • CThe model will automatically correct the biases during fine-tuning
   • DThe model may perpetuate or amplify those biases in its predictions
   Reveal Solution Discussion Select an answer first