How do autoencoders work for unsupervised representation learning?

Autoencoders compress input data into a latent space and reconstruct it back to learn meaningful representations.

Autoencoders consist of an encoder network that maps input data to a lower-dimensional latent representation, and a decoder network that reconstructs the original input. Training minimizes reconstruction loss (e.g., MSE). The bottleneck forces the model to learn compressed, meaningful features rather than memorizing input. Variants like denoising autoencoders, sparse autoencoders, and variational autoencoders improve robustness and probabilistic modeling.