Neurons and Neural Networks
Neurons essentially an IN / OUT function, the internal neurons often used sigmoid, tanh, ReLu the nonlinear function. A large neural networks are often made up of many neurons.
Sigmoid expressions and graphics
ReLu expressions and graphics
Supervised learning
During training, the use of the data is tagged, this training to learn the way it is called supervised learning. For example, there are a variety of images on the Web, now need to be able to identify training a cat model, we need to classify the picture will be placed in a pile of cat pictures, other images on another pile, so to pictures added this label is not part of the cat, then use a bunch of pictures that belong to train a cat model, eventually will be able to complete the task. Because in the learning process, the data used is tagged, so this approach to learning is also called supervised learning. So similar, in that the training data used are unlabeled.
Unsupervised Learning
Unsupervised learning is also similar to that in training data used are unlabeled. Now with the most depth learning the best thing supervised learning, and perhaps will become the mainstream in the future unsupervised learning.
Structured data and unstructured data
Generally refers to structured data based on data in a database, such as in a data table, the data structure of each example are the same.
Unstructured data typically refers to audio, image, video and other data.
CNN and RNN
CNN called the depth of Chinese neural network, it convolution structure on neural networks, can be good for image processing.
RNN Chinese called loop neural network, usually a processing sequence of data, such as audio, video and the like.