”“…… I hear birds singing everywhere… ” After a brief pause, the reporter recited the famous first sentence of the Tang poem “Spring Dawn” to a computer named “Darwin mouse”. After a short pause, it recited the whole poem with its own “memory”. This computer is the first brain like computer in China recently released by the United River Laboratory of Zhejiang University. Working like the brain neurons, it contains 120 million pulsing neurons and nearly 100 billion synapses. What is the great “super power” of this computer? < / P > < p > in a laboratory of Zhejiang University, three 1.6-meter-high standard server cabinets stand side by side, with red lights flashing inside the black enclosure. < / P > < p > “although the shape is no different from the general computer, the core of brain like computer is very different.” Pan Gang, head of the research team and professor of the school of computer science and technology of Zhejiang University, said that the brain like computer contains 792 Darwinian generation 2 brain like chips developed by Zhejiang University, and its typical operation power consumption is only 350-500 watts. The chip of brain like computer simulates the structure and function mechanism of brain neural network, and has obvious advantages in image, video and natural language fuzzy processing. Recently, it has become the competitive frontier of computer science to create such a computer. In 2015 and 2019, Zhejiang University has successfully developed Darwinian generation I and Darwinian II brain like computing chips. On this basis, the team is marching towards brain like computers. < / P > < p > “in short, the working mechanism of brain neurons can be understood as follows: as neurons receive input pulses, the membrane voltage of their cell bodies increases. When reaching a certain threshold, an output pulse will be sent to the axon and transmitted to adjacent neurons to realize the transmission of information. When the signal comes, it will start up, and it will not run if there is no signal. ” Ma De, an associate professor of Zhejiang University and the backbone of the project, said that the working principle of brain like computers is extracted from the “daily life” of biological neurons. Compared with traditional computers, the energy consumption is significantly reduced and the efficiency is greatly improved. < / P > < p > in order to make a large number of neurons interact efficiently and orderly distribute the disorderly information flow, researchers have also developed a Darwinian brain like operating system. At present, the function task switching time of the system has reached microseconds. Compared with the small and light electronic computers we use every day, today’s brain like computers are still “big”. But scientists say that if measured by the history of electronic computers, brain like computers are still in the “ENIAC” stage of all initial creation. With the continuous iterative upgrading of chips and other hardware, the “weight loss” of brain like computers is just around the corner. Traditionally, computers rely on the von Neumann architecture, which is better than numerical computation. However, with the gradual failure of Moore’s theorem, the limitations of this architecture are increasingly obvious. From storage to power consumption, to the level of intelligence, forward-looking scientists have seen its “bottleneck”. < / P > < p > “in the von Neumann architecture, data storage and calculation are separated. This is like information stored in a place, but when you want to calculate, you have to move the information to another place, and then move back after the calculation. But the speed of transportation is much lower than the calculated speed, which makes the transportation itself a trouble. ” Pan Gang said that the “storage wall” has made it difficult to improve the performance of computers in the era of big data. If you want to engage in high-energy computing such as artificial intelligence, computers will hit the “power wall”. < p > < p > at the moment when data has become a basic strategic resource and computing power has become an important measure of national comprehensive strength, how to realize disruptive innovation in computer development has become the “battlefield of strategists” in the scientific circles of various countries today. Learning from human brain has become the focus of exploration. The United States has invested US $100 million to support the research team of IBM and HRL to develop pseudo neural network computing; Heidelberg University and other universities in Germany have actively deployed brain computers Zhejiang University team is trying to catch up. < / P > < p > if there is an uninvited mouse in the room, it’s a headache. It may come and go without a trace, eat food without saying, and can accurately bypass the traps set by people. What level of “intelligence” can be achieved by a brain like computer with the same operation scale as mouse brain neurons? < / P > < p > in the experiment of Zhejiang University project, the brain like computer showed many abilities, such as directing the robot to complete the “flood fighting and disaster relief” simulation task. At the experimental site, banyuetan reporter saw that three robots perform their respective duties under the computer scheduling. Robot 1 carries its own camera and patrols the site. Once a gap is found, it calls robot 3, which is responsible for engineering maintenance, to repair the dam. After finding a human model that has fallen on the ground, he calls robot 2 responsible for rescue < / P > < p > “these robots are controlled by different” brain regions “of brain like computers, and their functions are not fixed. Robot No. 1 is now on patrol and can be responsible for rescue or engineering maintenance in a while. ” Project research backbone, Zhejiang University associate professor Li Ying said. < p > < p > “mind typing”, EEG real-time decoding, diagnosis of harmful air components All kinds of brain like computer performance of “learning language with one’s teeth” makes one’s eyes shine. However, Pan Gang also pointed out that the current brain like chip uses a simplified neuron model, while the behavior and connection mode of real brain neurons are too complex. At present, for scientists, the brain is still a “black box”, and many mechanisms remain unsolved. There is still a long way to go from understanding the brain world to creating new algorithms. Didi Qingju bicycle has entered 150 cities

By ibmwl