Task 2.2: Train and Refine Models MLA-C01 Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A team is training a neural network for image classification. The training loss decreases steadily, but the validation loss starts increasing after epoch 10 while training loss continues to drop. The team needs to improve generalization without significantly increasing training time. Which approach should they take?

   • AIncrease the number of epochs to 50 and keep the current architecture unchanged.
   • BReduce the training dataset size to make the model simpler.
   • CAdd dropout layers and L2 weight decay to the model, and retrain from scratch with early stopping monitored on validation loss.
   • DRemove the dropout layers to allow the model to fit the training data more closely.
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A company has multiple ML teams that train and deploy models. They need to maintain a clear record of model versions, their associated metrics, and approval status for audit purposes. Which SageMaker feature should they use?

   • ASageMaker Feature Store
   • BSageMaker Model Monitor
   • CSageMaker Experiments
   • DSageMaker Model Registry
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   Which SageMaker feature is used to automate the search for optimal hyperparameters?

   • AExperiments
   • BData Wrangler
   • CAutomatic Model Tuning
   • DModel Registry
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   Which approach is most directly aimed at reducing the wall-clock time required to train a large deep learning model?

   • AUsing a larger batch size without adjusting learning rate
   • BAdding more layers to the model
   • CDistributed training across multiple GPUs
   • DIncreasing the number of epochs
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A data science team is training a deep learning model on a large dataset using SageMaker. The training job is taking 6 hours to complete, and the team needs to reduce this time to under 2 hours to meet a project deadline. They have already confirmed that the model architecture and data pipeline are correct. Which combination of actions will most directly reduce the training wall-clock time?

   • ADecrease the batch size and increase the number of epochs to improve convergence accuracy.
   • BSwitch to a single larger GPU instance without changing the batch size or training script.
   • CIncrease the batch size and enable distributed training across multiple GPU instances using SageMaker's data parallelism library.
   • DEnable SageMaker Automatic Model Tuning to run a hyperparameter search job in parallel with the current training.
   Reveal Solution Discussion Select an answer first