Architectures · advanced · concept 103 of 311 · Updated Nov 4, 2026
Boltzmann Machines
A Boltzmann machine is a network that learns the probability of patterns in data using ideas from statistical physics. Restricted versions were key to the deep learning revival of the mid-2000s, because stacking them made training deep networks workable. Few people use them now, yet they help explain energy-based and generative models, and they feature in the 2024 Nobel Prize in Physics.
Key terms
Restricted Boltzmann machineEnergy functionContrastive divergenceDeep belief networkStochastic units
Learn these first
Where you meet it in the real world
Historical foundation of deep generative models; early recommendation systems
Guides and articles
Nobel Prize lecture ↗
Geoffrey Hinton