Several Basic Concepts of Neural Network (Deep Learning)

Broadly speaking, the network structure of deep learning is also a multi-layer neural network. In the traditional sense, multi-layer neural network only has input layer, hidden layer and output layer. Among them, the number of hidden layers depends on the need, and there is no clear theoretical derivation to explain how many layers are appropriate. CNN, the most famous convolutional neural network in deep learning, adds a feature learning part to the original multi-layer neural network, imitating the classification of signal processing by human brain. The specific operation is to add a partially connected convolution layer and a dimension reduction layer in front of the original fully connected layer, and add a layer. Input layer-convolution layer-dimensionality reduction layer-convolution layer-dimensionality reduction layer-hidden layer-output layer In short, the initial steps of multilayer neural network are: feature mapping to values. Characterized by manual selection. The step of deep learning is-> signal; Characteristics-> Value. Features are selected by the network itself.