Restricted Boltzmann machine
Máquina de Boltzmann restringida
- Date
- 1986 · confidence high
- Place
- Boulder, Colorado, United States
- Credited to
- Paul Smolensky y Geoffrey Hinton
- Field
- Knowledge & symbols
What it is
A generative stochastic neural network organized in two layers with no internal connections, able to learn a probability distribution over its inputs; proposed as «Harmonium» by Paul Smolensky in 1986 and popularized by Geoffrey Hinton with fast learning algorithms in the mid-2000s.
What it led to
- Basis for layer-by-layer pretraining of deep neural networks
- Boost to the resurgence of deep learning in the 2000s
- Application in recommendation systems and dimensionality reduction