148_1000.jpg -

Using a pre-trained ResNet-50 or Vision Transformer (ViT) to extract the embedding vector for 148_1000.jpg .

To investigate the representational value of specific data points within the broader training set. 2. Methodology 148_1000.jpg

Applying t-SNE or UMAP to see where this image sits relative to its assigned class. Using a pre-trained ResNet-50 or Vision Transformer (ViT)

Measuring the cross-entropy loss contribution of this single image during a training epoch. 148_1000.jpg