| 摘要 |
A conditional generative model for generating sample data or outputs similar to those that have been observed by conditionally using stochastic selection categorisation probability, i.e. choosing according to a probability that the output generated falls in to a specific category, then using a deterministic neural network to output the nearest previously seen example to the input. The neural network may be a multilayer perceptron and it may form a modified Helmholtz machine. The use of the neural network artificial intelligence classifier is termed signal processing using compressed mixture and chained compressed mixture approaches. The neural networks may be chained together so they operate sequentially. Potential applications include: simulating imagination, for example outputting example imagined data based on an input category; completion of partial signals, e.g. images in character recognition; and, classifying objects amongst categories; |
