AWS Certified Machine Learning Engineer Associate · Free Practice Question Hard
Question 25
A machine learning engineer is training a deep learning model on SageMaker using TensorFlow. During training, the model's performance stops improving, and the engineer suspects that the model may be experiencing vanishing gradients. They want to use SageMaker Debugger to monitor the training and resolve this issue. Which steps should the engineer take to identify and address the vanishing gradients using SageMaker Debugger?
-
A
Utilize SageMaker Profiler to collect GPU utilization data and implement early stopping when vanishing gradients are detected.
-
B
Run a custom SageMaker Debugger rule to visualize CPU and memory usage, then adjust the batch size to optimize resource utilization.
-
C
Use SageMaker Debugger’s built-in rules to monitor the gradient values during training and adjust the learning rate if vanishing gradients are detected.
-
D
Enable SageMaker Debugger to monitor resource utilization and set up alarms for underutilized GPUs.
Reveal correct answer
Correct answer: C
A.
The SageMaker Profiler focuses on resource utilization (e.g., CPU/GPU), not model training metrics like gradients, so it wouldn’t help detect vanishing gradients.
B.
Visualizing CPU and memory usage does not directly address vanishing gradients, which are a model training issue.
C.
SageMaker Debugger provides real-time monitoring for training jobs and includes built-in rules to detect vanishing gradients, which can hinder model improvement. The engineer can adjust learning rate or other hyperparameters when such an issue is detected, improving training results.
D.
While SageMaker Debugger can monitor resources, vanishing gradients are a model-specific issue related to learning dynamics, not GPU utilization.
Discussion
Think the marked answer is wrong, or have a better explanation? Share it below — comments appear after review.
