The History of AI

The Boltzmann machine

Era 3 · From the expert boom to the statistical turn (1980–2011) › Neural networks

Lineage
Neurons and networks
Kind
Publication

Who

Carnegie Mellon, Johns Hopkins University

What happened

David Ackley, Geoffrey Hinton, and Terrence Sejnowski publish "A Learning Algorithm for Boltzmann Machines" in Cognitive Science, a stochastic network that learns the probability distribution of its data.

Why it mattered

One of the first networks able to learn internal representations in hidden units. It brings statistical physics to learning and stands at the origin of generative models.

The full story

The Hopfield network stored patterns but did not learn to discover them. Hinton and Sejnowski set out to build a network that, from examples, would construct its own representation of their structure. To do so they added hidden units, which correspond to no data, and replaced deterministic decisions with random ones.

In a Boltzmann machine, each unit switches on with a probability that depends on its inputs and on a "temperature," like the atoms of a gas under the Boltzmann distribution of statistical physics. Left to run, the network visits states with a probability that depends on their energy. Learning adjusts the connections so that the states the network visits on its own resemble the training data. The resulting rule is local and elegant: it compares how two units are correlated when the network sees the data and when it runs freely.

The method was very slow, because the network had to be allowed to reach equilibrium again and again. But it showed that a network could learn useful internal representations from a clear principle, and its restricted version, with no connections within each layer, proposed by Paul Smolensky in 1986 and which Hinton learned to train quickly in 2002, became the building block of deep belief networks.

The Nobel committee cited the Boltzmann machine in awarding Hinton the 2024 Nobel Prize in Physics.

Builds on

  • The Hopfield network (April 1982)

    Documented direct influence

    The Nobel committee explains that Hinton used the Hopfield network as the basis for a new network, the Boltzmann machine, with tools from statistical physics

  • The Hopfield network (April 1982)

    Documented direct influence

    Hinton builds on the Hopfield network to create the Boltzmann machine, with tools from statistical physics

Leads to

  • Deep belief nets (July 2006)

    Documented direct influence

    Each layer of the deep belief network is a restricted Boltzmann machine trained on its own

  • Nobel Prize in Physics for Hopfield and Hinton (October 8, 2024)

    Documented direct influence

    The Academy's announcement names the Hopfield network and the Boltzmann machine as the work being honored

Threads that pass through here

Sources

How to cite

“The Boltzmann machine.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/boltzmann-machine

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline