SIGNAL//DESK
AI/MLsrc: curated AI glossary

Unsupervised learning

Unsupervised learning is like giving a computer a giant pile of photos and letting it sort them into groups on its own, without telling it what is in the pictures or what to look for.

A machine learning approach where models identify inherent patterns, clusters, or latent structures within datasets that lack ground-truth labels, typically utilized for tasks like clustering, anomaly detection, and dimensionality reduction.

A learning paradigm characterized by the optimization of an objective function based solely on the intrinsic properties of the input distribution, enabling the discovery of manifold structures, density estimations, or latent representations in the absence of explicit supervisory signals or target variables.

evolution

  1. 1950 · history
    Hebbian Theory

    Donald Hebb proposed a neurophysiological basis for unsupervised learning through synaptic plasticity.

  2. 1982 · history
    Hopfield Networks

    John Hopfield introduced recurrent neural networks capable of associative memory and pattern reconstruction without explicit labels.

  3. 1986 · history
    Boltzmann Machines

    Hinton and Sejnowski developed stochastic recurrent networks that learn internal representations of data distributions.

  4. 2006 · history
    Deep Belief Networks

    Geoffrey Hinton popularized greedy layer-wise pre-training, enabling the effective training of deep unsupervised architectures.

  5. 2014 · history
    Generative Adversarial Networks

    Ian Goodfellow introduced GANs, revolutionizing unsupervised generative modeling through a competitive game-theoretic framework.


← all terms