Exam AWS Certified Machine Learning - Specialty topic 1 question 265 discussion

My PoV is that D requires LEAST amount of effort from the internal team. The blog post (https://aws.amazon.com/blogs/machine-learning/identify-rooftop-solar-panels-from-satellite-imagery-using-amazon-rekognition-custom-labels/) does not mention that the effort would be huge, to label 8k images even with Ground Truth active learning. Far more than the effort of using SageMaker Object Detection algorithm and SageMaker batch transform. Considering the fact that the internal team has no ML expertise and no ML experience, they need to do some configuration, learn key concepts (e.g., preparing data in the correct format, specifying hyperparameters), and follow AWS documentation and industrial best practices for training and inference...

https://aws.amazon.com/blogs/machine-learning/identify-rooftop-solar-panels-from-satellite-imagery-using-amazon-rekognition-custom-labels/

Least effort + no ML experience, so A

A. YES - Amazon Rekognition Custom Labels is better than other option like Face/Celebrity/etc.; Ground Truth Active Learning will require human labelling only when needed, works well with small internal team B. NO - missing Active Learning C. NO - SageMaker Object Detection is more complicated than labelling D. NO

Vote for A

SageMaker Ground Truth can use active learning to automate the labeling of the input data for certain built-in task types, such as object detection. Active learning is a machine learning technique that identifies data that should be labeled by human workers. This helps to reduce the cost and time that it takes to label the dataset compared to using only humans1. By setting up a private workforce, the internal team can use their own domain knowledge to label the data and ensure quality and consistency.

As we have an internal team working on this project, it is understood that they will do the labeling. Letter A is correct, as SageMaker Active Learning Feature allows you to streamline the team's efforts. Letters C - D are wrong as they use the wrong algorithm (object detection) and Letter B takes longer than Letter A.

Option D uses a public workforce to label the data. This means that the company can leverage a large pool of workers from Amazon Mechanical Turk, who are experienced and qualified in labeling tasks. The public workforce can provide more diverse and accurate labels than the internal team, who may have limited or biased perspectives. The public workforce can also complete the labeling task faster and more efficiently than the internal team, who may have other priorities or responsibilities.

https://aws.amazon.com/blogs/machine-learning/identify-rooftop-solar-panels-from-satellite-imagery-using-amazon-rekognition-custom-labels/

https://docs.aws.amazon.com/sagemaker/latest/dg/sms-automated-labeling.html

It's A due to small team on the project and minimal effort from the team required. SageMaker Ground Truth active learning feature can speed up the labeling process for 8000 images.

B is correct

B. Set up a private workforce that consists of the internal team. Use the private workforce to label the data. Use Amazon Rekognition Custom Labels for model training and hosting. By setting up a private workforce consisting of the internal team and using Amazon Rekognition Custom Labels, the company can leverage the labeling capabilities of the internal team to label the data. Amazon Rekognition Custom Labels can then be used for model training and hosting. This option eliminates the need for additional complex steps such as active learning or object detection algorithm training, which may require more ML expertise and effort from the internal team. Instead, it relies on the simplicity and convenience of using Amazon Rekognition Custom Labels for model training and hosting, making it the least effort-intensive option for the team with no ML expertise or experience.

C is correct.