Loading filters...
Wake Sleep

Wake Sleep

Biologically inspired algorithm used within unsupervised learning to train deep belief networks.

The Wake-Sleep algorithm is a method used within unsupervised learning, specifically for training generative models such as deep belief networks. It works by simulating two stages: a 'wake' phase, where the system propagates in the forward direction and adjusts the recognition weights, and a 'sleep' phase, where the system propagates in the reverse direction adjusting the generative weights. It is particularly significant within AI because it provides a potential solution for adapting the weights in multilayer neural networks, a problem that is central to deep learning. The algorithm can be used in a variety of AI applications, such as speech recognition, image recognition, and natural language processing.

The Wake-Sleep algorithm was introduced in 1995 by Geoffrey Hinton, Peter Dayan, Brendan Frey, and Radford Neal. It was part of the wave of research in the 1990s that led to the rediscovery of backpropagation and the development of deep learning.

The main contributors to the development and advancement of the Wake-Sleep algorithm are Geoffrey Hinton, Peter Dayan, Brendan Frey, and Radford Neal. Geoffrey Hinton, known as the 'godfather of deep learning,' has made numerous significant contributions to the field of AI, including this work on the Wake-Sleep algorithm. Peter Dayan is a renowned computational neuroscientist, and both Brendan Frey and Radford Neal are leading researchers in machine learning.

Generality: 0.54