Which actions should the ML specialist take to address this problem?
(Choose two.)
Use Amazon SageMaker Ground Truth to label the unlabeled images.
Use image preprocessing to transform the images into grayscale images.
Use data augmentation to rotate and translate the labeled images.
Replace the activation of the last layer with a sigmoid.
Use the Amazon SageMaker k-nearest neighbors (k-NN) algorithm to label the unlabeled images.
Explanations:
Using Amazon SageMaker Ground Truth to label the unlabeled images will help increase the size and diversity of the labeled dataset, which can improve generalization and reduce overfitting.
Converting the images to grayscale could remove useful color information, which might hurt model performance and does not address overfitting directly.
Data augmentation, such as rotating and translating the images, can artificially increase the variety of the training data, helping to prevent overfitting by promoting model generalization.
Replacing the activation of the last layer with a sigmoid is not appropriate for a multi-class classification problem. A softmax activation is better suited for this task.
The k-NN algorithm is not typically used for labeling images in an active learning context. Labeling images using k-NN would require manually selected labeled examples and may not be the most efficient approach.
I suspect that the answer is:
Use Amazon SageMaker Ground Truth to label the unlabeled images.