VAE
A Variational Autoencoder is a generative model that learns to compress data into a structured probabilistic latent space from which new data is sampled.
Think of It Like This
Like a compression program that doesn't just store files, but learns the 'idea' of the files so it can randomly generate brand-new, similar ones.
Unlike standard autoencoders that map inputs to strict points, VAEs map inputs to probability distributions (means and variances). The model is forced to reconstruct the input by sampling from these distributions, regulated by the KL divergence. This creates a smooth, continuous latent space perfectly suited for generative tasks.