restricted Boltzmann machine

Terms from Artificial Intelligence: humans at the heart of algorithms

The glossary is being gradually proof checked, but currently has many typos and misspellings.

A restricted Boltzmann machine (RBM) is a form of Boltzmann machine where input and output layers are fully connected to each another, but have no internal connections. That is there are input-ouptut connections, but neither input-input, nor output-output links. Otherwise they are like a standard Boltzmann machine with a training phase with inputs clamped and outputs free to vary and then a recall mode where some inputs are clamped and the network fills in the gaps.
As well as this reconstruction mode, RBMs are often used for feature detection in the early stages of deep neural networks as they use feedforward learning rather than backpropagation.
The restricted Boltzmann machine, like the Boltzmann machine, is a form of neural network and autoencoder.

Used in Chap. 6: page 81; Chap. 8: page 104; Chap. 10: page 138; Chap. 12: page 182

Also known as RBM

Restricted Boltzmann Machine