Latent Space

The abstract, lower-dimensional representation space learned by neural networks to encode the essential features of input data. In generative models, points in latent space correspond to potential outputs — images, text, or audio. Navigating latent space enables interpolation between concepts and controlled generation.