These days, Machine Learning, as an important branch of computer science, suddenly bursts into our vision. To support machine learning, Deep Learning, a cutting edge technique, brings lots of possibilities to extend artificial intelligence area. Deep Learning has been used in many areas of our daily live. From Alpha-Go to AI robot Sophia, artificial intelligence not only brings excitements and future , but also brings nervous and threats to human. Deep learning is not just an algorithm or research project on papers. It has been applied to real physical world, especially some highly human involving systems. Hacking into those human involving systems would not only cause computer shutting down or rebooting system, but will effect human live. In software testing, we called this {it hazard}. Most resent hot anti-deep learning security topic is {f Adversarial Examples}. In order to understand the adversarial examples, we need to at least have a concept about deep learning. How it works? What is the architecture and the technique underneath the hood?{it Deep Neural Networks } (DNNs) is a sequence of computer algorithms imitating how human brain thinking and perceiving the real world. The hierarchy of deep neural networks simulates the layer of human. Intuitively, we know that our brains consist with many different cells. The most important one is called neuron, which helps us recognize, understand everything surrender us. Millions of neurons build deferent level of layers. Connecting those layers leads to a powerful network consisting our brain. This brilliant network is the prototype of DNNs. In DNNs, each calculation unit is called artificial neuron having the same functionality as really human neuron. These artificial neurons receive data from other neurons. Under some calculations, the neurons pass generating data to neurons lay on the next layer. Though multiple layers, we can get useful information that we want. the visualized topology of deep neural net works shows in Fig. 1.