Task 2.2: Train and Refine Models MLA-C01 Practice Questions (Page 8)

Questions 36–40

1. 36foundation · easy

   Which AWS service provides pre-trained foundation models that can be fine-tuned with custom datasets?

   • AAmazon Bedrock
   • BAmazon S3
   • CAmazon QuickSight
   • DAWS Lambda
   Reveal Solution Discussion Select an answer first
2. 37foundation · easy

   A training run uses a dataset of 1,000 samples with a batch size of 100. How many steps are in one epoch?

   • A1,000
   • B10
   • C100
   • D1
   Reveal Solution Discussion Select an answer first
4. 38foundation · easy

   What is the main idea behind ensemble learning?

   • ACombining multiple models to improve overall performance.
   • BTraining a single model on multiple datasets.
   • CUsing a larger model to capture more patterns.
   • DSelecting the best model from a set of candidates.
   Reveal Solution Discussion Select an answer first
5. 39application · medium

   A team has trained a large image classification model that achieves 95% accuracy. They need to deploy it to edge devices with limited memory and storage. The model is currently 500 MB and the device has only 100 MB of available storage. The team must reduce the model size while minimizing accuracy loss. Which approach should they use?

   • ADeploy the model as-is and rely on the device's virtual memory to handle the size.
   • BApply post-training quantization to convert the model weights from FP32 to INT8.
   • CUse feature selection to remove less important input features from the model.
   • DIncrease the number of layers in the model to improve accuracy before deployment.
   Reveal Solution Discussion Select an answer first
6. 40expert · hard

   A team is fine-tuning a large pre-trained language model on a small domain-specific dataset. The model performs well on the training data but poorly on the validation set. The team suspects overfitting and also wants to avoid catastrophic forgetting of the pre-trained knowledge. Which combination of techniques should they apply?

   • AUse a higher learning rate to adapt the model faster to the new domain.
   • BUse a lower learning rate, add L2 regularization, and use early stopping on the validation loss.
   • CTrain the model for more epochs to ensure it fully learns the domain-specific patterns.
   • DFreeze all layers of the pre-trained model and only train a new classification head.
   Reveal Solution Discussion Select an answer first