It refers to the neural network ANN or called connection model, which is a feature of neural networks mimic the behavior of animals, the algorithm mathematical model of distributed parallel processing of information. Such networks rely on complexity of the system, by adjusting the relationship between the internal large number of interconnected nodes, so as to achieve the purpose of processing information. Neural network algorithm used in vector multiplication using sign function and its various approximation. Parallel, fault tolerance, and self-learning can be implemented in hardware features, a few basic advantages of neural networks, where is the difference between the calculation method and the traditional method of neural networks.
Learning from the concept of the depth of artificial neural networks, multilayer perceptron containing multiple hidden layers is a deep learning architecture. Attribute category indicates deep learning or more abstract features formed by the combination of high-level low-level features, to find a distributed representation of the characteristic data. Convinced of the proposed network based unsupervised training algorithm greedy layer by layer, bring hope to solve the optimization problem of deep structure, and then put forward a multi-layer automatic encoder deep structure.
Multilayer neural network in the traditional sense only input layer, hidden layer and output layer, the number of layers in which the hidden layer if necessary may be, there is no clear theoretical analysis to explain in the end how many layers appropriate steps multilayer neural network to do is: mapped to a value characteristic, characterized in that the artificial selection.
A network structure of said depth study of multilayer neural networks also broadly. The most famous of convolution depth learning neural network by Lecun proposed by others, is the first truly multi-layer structure learning algorithm, which uses spatially relative reduction in the number of parameters to improve training performance. On the basis of the original multilayer neural networks, adding the characteristics of a learning section, which is modeled after the human brain graded on signal processing. Specific operations in the front layer is connected to the original full convolution added dimensionality reduction layer and layer connection portion, and is added to a level: input layer - layer Convolution - dimensionality reduction layer - layer Convolution - dimensionality reduction layer - ....- hidden layer - the output layer. The depth of the step is to learn to do: Signal -> Characteristics -> value, characterized in that the selection by the network itself.