Dimensionality Reduction
The goal of this method is to visualize the distance between feature vectors with high dimensions into the form which we, the human, can understand
This method first records the features vectors in FC layer in CNNs, then reduce the dimensionality of vectors to 2 dimensions with algorithms
Common Algorithms:
- Simple: Principal Component Analysis (PCA)
- Complex: t-SNE
Two images which are near to each other in the picture below have similar feature maps
