An ML specialist is building a computer vision model to classify 10 traffic sign categories. The team has 100 labeled images per class stored in Amazon S3 and an additional 10,000 unlabeled images. All images are 224×224 from dash cameras. After several training runs the model is overfitting the training data. Which actions should the specialist take to reduce overfitting? (Choose two.)

Overfitting occurs when a model learns the training data too well, including noise, and performs poorly on unseen data. Using Amazon SageMaker Ground Truth to label the unlabeled images directly addresses overfitting by increasing the size and diversity of the training dataset. A larger, more representative dataset helps the model generalize better. Applying data augmentation (e.g., rotations, translations) to the labeled images artificially expands the training dataset by creating new, slightly modified versions of existing images. This exposes the model to more variations of the data, making it more robust and less likely to memorize specific training examples. Converting images to grayscale reduces information and may hinder classification performance, not necessarily reduce overfitting. Changing the activation function of the final layer to a sigmoid is incorrect because sigmoid is typically used for binary classification, not multi-class classification (10 categories). The k-nearest neighbors (k-NN) algorithm is not ideal for labeling a large dataset of images and is unlikely to be as effective or scalable as Ground Truth for this task.