LAUSR

Deep learning is a model with multi-level layer structure that uses the underlying output as input from the top. From down to above is a process of the unsupervised learning, which automatically learns useful features, and expresses the low-level features as advanced features and from top to bottom is supervised learning process that through the labeled data to the whole network parameter optimization and adjustment of the whole network which has the characteristics of better learning ability. Deep learning has been able to develop so rapidly in recent years mainly due to the following two reasons. (1) The application of massive tagged data mitigates the problem of training. In deep learning, the data is "engine", and Imagenet has millions of annotated data. (2) The rapid development of computer hardware provides a powerful computing power which makes it possible to train large-scale neural networks, such as high-performance GPU can integrate thousands of cores.