The Boltzmann machine

Author

Thiam, Tafsir

Author_Institution

Woodrow Wilson SHS, Washington, DC, USA

Volume

6

fYear

1999

fDate

36342

Firstpage

4428

Abstract

Boltzmann machines are discrete networks with input, output and hidden units, in which the units update their state with a stochastic function. The output of a given node is calculated using probabilities, rather than threshold or sigmoid function. The paper describes their basic architecture, their processing mechanism, the principles of their operation, and their learning mechanism. The main characteristics of the Boltzmann machine is the fact that, when subjected to reducing noise, it has a final probability of resting in given states which is in direct proportion to Hopfield´s calculation of the energy of those states. The key problem is to control the values of such energies by changing the weights

Keywords

Boltzmann machines; learning (artificial intelligence); multilayer perceptrons; noise; probability; stochastic processes; Boltzmann machine; discrete networks; learning mechanism; neural net architecture; probabilities; processing mechanism; reducing noise; state updating; stochastic function; Algorithm design and analysis; Annealing; Energy states; Machine learning; Temperature; Transfer functions;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks, 1999. IJCNN '99. International Joint Conference on

Conference_Location

Washington, DC

ISSN

1098-7576

Print_ISBN

0-7803-5529-6

Type

conf

DOI

10.1109/IJCNN.1999.830883

Filename

830883