The essence of the human brain is a network composed of neurons. The scientific community generally believes that the human brain has 100 billion neurons. The basic structure of neurons is shown in Figure 1. Each neuron has a large number of nerve fibers. Among these nerve fibers, dendrites account for the majority and only one axon (but can be bifurcated). Dendrites are responsible for receiving and transmitting information, and axons are responsible for outputting information. When the dendrites receive information greater than the excitability threshold, the entire neuron will generate a short but extremely obvious "action potential", which will spread throughout the entire neuron almost instantaneously, including the end of the nerve fiber away from the cell body. After that, the end structure of the "synapse" between the axon of the previous neuron and the dendrite of the next neuron will be activated by electrical signals, and the "neurotransmitter" will be released by the presynaptic membrane immediately. To transmit information between two neurons, and play a role in excitement or inhibition to the next neuron.

Figure 1 The basic structure of neurons

Since the human brain has functions such as large-scale parallel storage and processing, high-intelligence logical reasoning, and self-learning, people hope to apply the functions of the human brain to computers, so that it has the ability of human brain-like logical reasoning and self-learning. Artificial Intelligence (AI) is a new science that studies and develops theories, methods, technologies, and application systems used to simulate and extend human intelligence. The purpose of artificial intelligence is to make machines capable of complex tasks that require human intelligence, such as speech recognition, natural language processing, and robots. These artificial intelligence technologies are generally implemented through neural networks. Neural networks can approximate any continuous function infinitely, which is the main reason why scholars believe that it plays an important role in artificial intelligence. At present, artificial intelligence research has become more and more in-depth, and the role of neural networks has become increasingly prominent.

Artificial neural networks in the field of artificial intelligence are based on the structure of biological neural networks, mathematically abstract and simulate some of their basic characteristics, and are used to explore bionic models of human brain intelligent behavior. The emergence and development of artificial neural networks are closely related to computer science, artificial intelligence, and neuroscience. At present, artificial neural networks have made some achievements in practical applications such as pattern recognition, information processing, and logical operations. Therefore, the research on artificial neural networks has gradually become a hot topic and has attracted the attention of a large number of scholars. In the 1940s, research scholars first began to study artificial neural networks. A neural network model usually consists of self-feedback connection weights, interconnection weights, and activation functions. The first mathematical model of artificial neural network was the MP mathematical model jointly proposed by W.S. McCulloch and W. Pitts in 1943. The MP model proved that a single neuron has the ability to process logical operations, and has since created the era of artificial neural network research. D.O. Hebb proposed a learning rule in 1949, which believes that the size of the connection strength between neurons can be changed. In the 1960s, F. Rosenblatt designed the perceptron. Compared with MP model and Hebb rule, the perceptron model is a more complete neural network model. The perceptron model can store the information in the neural network in the connection weights between neurons. M. Minsky and S. Papert wrote a book called "Perception" in 1969, which theoretically proved that the perceptron model's ability to process information is limited, including the inability to implement XOR operations and predicate operations. So after more than ten years, the research in the field of neural networks entered a low ebb.

Until 1982, J.J. Hopfield proposed a new network model, which is the well-known Hopfield neural network model. The proposal of the Hopfield neural network model has strongly promoted the development of neural computing, and has triggered an upsurge of scholars on neural networks and related aspects. Equation (1.1) is the mathematical description of the neural network model implemented by J.J. Hopfield with the circuit

Among them, um represents the average somatic potential of the neuron, Cm and Rm are the input capacitance and resistance, respectively, and represent the finite impedance between the mth neuron and the output Vz, Vm=gm(um), gm(um) represents the response The input and output characteristics of the nonlinear amplifier with negligible time, Im is the input current of the mth neuron.

Since then, researchers have proposed several important neural network models based on the Hopfield neural network model. For example, M.A. Cohen and S. Grossberg proposed the Cohen-Grossberg neural network model in 1983. In 1988, L.O. Chua and L. Yang proposed a cellular neural network model.

Virtual reality (VR), also known as virtual technology, is the use of computer simulation to generate a virtual world in a three-dimensional space, providing users with a simulation of vision and other senses, making users feel as if they are on the scene, and can observe three-dimensional space instantly and without restrictions Things within. There is no doubt that artificial intelligence and virtual reality are the two most popular words in Silicon Valley today. Neural networks are a form of artificial intelligence. A large number of Internet devices are like neuronal networks in the human brain. They can search for deforestation; they can also track global crops to analyze future food shortages; they can also Monitor global oil tankers, predict future natural gas, etc.

Imagine what it would be like to connect a (artificial) neural network to virtual reality. For example, by wearing a headset that transmits brain waves to control the actions and expressions of characters in virtual reality; using artificial neural networks to quickly create 3D objects to achieve the required virtual reality world, when we wear the virtual reality device, in a way of crossing Revisit the scenes and life of childhood hometown and so on.

Before realizing (artificial) neural network access to virtual reality, I think the following two problems need to be solved.

1. The acquisition of human emotions, that is, how the head-mounted device obtains accurate brain waves to reflect the positive or negative emotional response of the user to the stimulation of the virtual reality scene;

2. How does the headset and the virtual reality device work together? This requires that the headset can read and record the radio wave activity of neurons in time and feed it back to the virtual reality device, and it also requires the virtual reality device to be fast according to the scene imagined by the human brain Construct a corresponding virtual world;

Currently, there is no mature solution for neural network access to virtual reality, but with the rapid development of artificial intelligence and computing science, it is believed that a large number of related products will emerge in the near future.