Create useful latent spaces with embeddings, autoencoding objectives, contrastive learning, and dimensionality reduction. Visualize representations and measure whether semantic relationships are preserved.
