AWS Certified Machine Learning – Specialty — Question 276

A company's machine learning (ML) specialist is building a computer vision model to classify 10 different traffic signs. The company has stored 100 images of each class in Amazon S3, and the company has another 10,000 unlabeled images. All the images come from dash cameras and are a size of 224 pixels × 224 pixels. After several training runs, the model is overfitting on the training data.

Which actions should the ML specialist take to address this problem? (Choose two.)

Answer options

• A. Use Amazon SageMaker Ground Truth to label the unlabeled images.
• B. Use image preprocessing to transform the images into grayscale images.
• C. Use data augmentation to rotate and translate the labeled images.
• D. Replace the activation of the last layer with a sigmoid.
• E. Use the Amazon SageMaker k-nearest neighbors (k-NN) algorithm to label the unlabeled images.

Correct answer: A, C

Explanation

Overfitting occurs when a model learns the training data too well due to a small dataset size. Labeling the 10,000 unlabeled images using Amazon SageMaker Ground Truth (Option A) provides a significantly larger training set, while data augmentation techniques like rotation and translation (Option C) artificially increase dataset variety and prevent the model from memorizing specific orientations. Converting to grayscale (Option B) removes color information which is crucial for traffic signs, and a sigmoid activation (Option D) is for binary classification rather than this 10-class multi-class problem.