AIF-C01 Fundamentals of AI and ML Practice Question

A data scientist is using Amazon SageMaker to train a deep learning model for image classification. The training job is taking too long. The dataset consists of 100,000 images stored in Amazon S3. Which action can the data scientist take to reduce training time without modifying the model architecture?

⚠ Common exam trap

Watch out — candidates often confuse checkpointing (which helps with recovery, not speed) or reducing epochs (which changes training duration but also model performance) with legitimate performance optimizations, while overlooking that GPU acceleration directly addresses the computational bottleneck without altering the model or dataset.

Answer choices

Why each option matters

Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.

Correct answer & explanation

✓

Use a GPU instance type for training.

GPU instances are specifically designed for parallel processing of matrix operations, which are fundamental to deep learning training. By switching to a GPU instance type (e.g., p3 or p4d families) in SageMaker, the data scientist can significantly accelerate the training of the image classification model without altering the model architecture, as the dataset of 100,000 images benefits from GPU's massive parallelism for forward and backward passes.

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

✗
Convert images to CSV format before training.
Why it's wrong here
CSV conversion discards spatial pixel structure that convolutional layers require, so the image classifier cannot learn. It is tempting because CSV suits tabular data, and converting to CSV would be the correct choice for a SageMaker tabular regression or classification job on structured records.
✓
Use a GPU instance type for training.
Why this is correct
GPU instances accelerate the matrix operations in deep neural network training far more than CPU instances, cutting training time without altering the architecture. The 100,000 images in Amazon S3 already stream efficiently, so compute acceleration is the binding constraint.
✗
Enable checkpointing to save intermediate models.
Why it's wrong here
Checkpointing saves intermediate model states for resumption after failure; it does not reduce the compute time of a successful training run. It is tempting because checkpointing improves resilience, which is its actual purpose, not speed.
✗
Reduce the number of training epochs.
Why it's wrong here
Fewer epochs cut training time but halt before the model converges, degrading accuracy on the 100,000-image dataset. It is tempting because epoch count directly controls training duration, and reducing it is the correct choice when a model is overfitting and validation loss has already begun rising.

Quick reference

AWS S3 Storage Class Comparison

Storage Class	Min Duration	Retrieval	Use Case
S3 Standard	None	Immediate	Frequently accessed data
S3 Standard-IA	30 days	Immediate	Infrequent access, rapid retrieval
S3 One Zone-IA	30 days	Immediate	Non-critical infrequent data
S3 Intelligent-Tiering	None	Immediate–hours	Unknown or changing access patterns
S3 Glacier Instant	90 days	Milliseconds	Archive with instant retrieval
S3 Glacier Flexible	90 days	Minutes–hours	Archive, flexible retrieval
S3 Glacier Deep Archive	180 days	Hours	Long-term compliance archive

Go deeper

AIF-C01 Fundamentals of AI and ML Practice Question

Why each option matters

Option-by-option breakdown

AWS S3 Storage Class Comparison

Related to this question