Great! China has successfully developed a 100 million level neuron like brain computer

category:Internet
 Great! China has successfully developed a 100 million level neuron like brain computer


Press conference of major achievements

Three standard cabinets with a height of 1.6 meters stand side by side. The black shell gives people a cool feeling. The red signal lamp keeps flashing. If you get closer, it seems that you can hear the pulse signal running fast inside.

Brain like computer

Recently, Zhejiang University and Zhijiang laboratory jointly developed Chinas first Darwin mouse based on brain chip with independent intellectual property rights.

This brain like computer contains 792 Darwinian second generation brain like chips developed by Zhejiang University. It supports 120 million pulse neurons and nearly 100 billion synapses. It is similar to the number of neurons in mouse brain. The typical operation power consumption is only 350-500 watts. At the same time, it is the largest brain like computer in the world.

At the same time, the team also developed a brain like computer operating system, Darwin OS, to effectively manage and schedule the hardware resources of brain like computers and support the operation and application of brain like computers.

Brain like computer

Subverting the traditional new computing model

For todays computers, which have long been commonplace in various fields of work and life, perhaps we have forgotten that scientists originally wanted to simulate a human brain through a machine.

However, with the development of computer, the von Neumann architecture, which is good at numerical calculation, was chosen at that time, that is, digital addition, subtraction, multiplication and division was used to construct information architecture. With the gradual failure of Moores theorem, the limitations brought by von Neumann architecture are increasingly obvious. The problems of storage wall, power consumption wall and intelligent improvement make the current computer development face great challenges.

For example, the storage wall problem is caused by the separation of data storage and computing in the existing von Neumann architecture. Its like information is stored in place A. when you want to calculate, you move the information to location B, and then move it back to place a after calculation. But the speed of transportation is much lower than the speed of calculation, which makes the transportation itself a key bottleneck Pan Gang, head of the research team and professor of the school of computer science and technology of Zhejiang University, said that this kind of computing mode restricts the improvement of computing performance represented by big data. The resulting data running and high energy consumption calculation such as artificial intelligence make the power wall problem emerge. At the same time, data-driven intelligent algorithms and training require massive samples and intensive computing, but high-level abilities such as drawing inferences from one instance and self-learning are relatively poor. The current machine intelligence is far from human intelligence.

The researchers are talking about it

How to break through the bottleneck caused by the existing computing mode?

Biological brain can naturally produce different intelligent behaviors in the process of interaction with environment, including speech understanding, visual recognition, decision-making tasks, operation control, etc., and the energy consumption is very low. In nature, many insects with nerve elements less than 1 million can achieve real-time target tracking, path planning, navigation and obstacle avoidance.

Brain like computer application demonstration: olfactory recognition

Pan Gang said that the structure and operation mechanism of brain neural network can be simulated by hardware and software to construct a new artificial intelligence system. This new computing mode that subverts the traditional computing architecture is brain like computing. It is characterized by the integration of storage and computing, event driven and highly parallel. It is the research focus of international academic and industrial circles, and is also an important science and technology strategy. Brain like computing has been regarded as one of the important ways to solve the computational problems such as artificial intelligence.

In recent years, Zhejiang University has focused on the core fields of human intelligence and machine intelligence, and has implemented the convergence research program of brain science and artificial intelligence, which is referred to as the dual brain project. It hopes to learn from the structural model and functional mechanism of the brain, apply the cutting-edge achievements of brain science to artificial intelligence and other research fields, and establish a new computer architecture leading the future.

The important point here is to run asynchronously, that is, when the signal comes, it will not run if there is no signal. The working principle of brain like chip is similar to the behavior of neurons in biology. It transmits signals through pulses, which can achieve high parallelism and improve efficiency.

Brain computer interaction demonstration

With hardware, there has to be software.

Its easy to say, but its not so simple to make so many neurons interconnected and expand to achieve efficient linkage combination, and to orderly distribute the disordered information flow to the corresponding functional brain regions.

For this reason, researchers have developed a brain like operating system for brain like computers, Darwin OS.

Brain like computer application demonstration: multi robot cooperation in flood fighting

Brain like computer application demonstration: multi robot cooperation in flood fighting

It doesnt seem new, and existing robots can do it. But the biggest difference is that these robots are controlled by brain like computers to carry out mobile instructions through voice and receive task assignment. The tasks of different robots can be switched by command, that is to say, their functions are not fixed, but controlled by different brain regions. Robot 1 now works as a patrol, and later it can become responsible for rescue or engineering. Associate Professor Li Ying, the backbone of project research, said.

In another experimental scene, the team members sing two sentences of a song to the computer, and then the computer can sing the content of the follow-up song by recalling it.

This is a brain like computer that can access the memory information inside the brain by simulating the memory mechanism of the hippocampus, which is different from the retrieval function commonly used by us. Professor Tang Huajin, the first mock exam leader, said that DarwinMouse brain computer can learn the memory learning function of hippocampus by learning from the structure of hippocampus network and neural mechanism. By impressing the pulse encoding of memory, the same model can learn and memorizing different types of data such as voice, song and text.

Brain like computer application demonstration: memory recall of simulated hippocampus

The worlds first computer, which was born in 1946, weighed 28 tons and operated at a speed of 5000 addition operations per second. However, in the following 70 years, computer technology developed rapidly. The speed of brain like computers is likely to be surprising, too.

Despite the fact that todays brain like computers are bulky, scientists say that with the continuous iterative upgrading of Darwin chips and other hardware, the size reduction will be in the near future. In the future, brain like computers may be implanted into mobile phones and robots to generate new intelligent service experience.

Compared with the hardware update, how to make brain like computers more intelligent is the focus of scientists next research.

At present, the input signal of sensors on the market is still mainly digital. When applied to a Darwin mouse brain computer, a coding layer should be added to convert the signal into pulse type. In this process, the information will be lost and damaged, which will reduce the efficiency of the computer to a certain extent. If we can solve this problem, brain like computers will become more intelligent.

At present, the research on brain like computing is still in its infancy. Darwinmouse brain like computer is still far from the real human brain in terms of scale and intelligence. However, its significance lies in providing an important practical example for this technology path, providing a tool and platform for researchers to verify brain like algorithm, so as to be more robust and real-time Sex and intelligence to solve practical tasks.

Group photo of research team

With the development of neuroscience and the maturity of system software, tool chain and algorithm of brain like computer, the goal of researchers in hezhijiang Laboratory of Zhejiang University is to make brain like computer as universal as von Neumann architecture computer, work as efficiently as brain, coexist and complement with von Neumann architecture to solve different problems.