Talking about intelligent control method , Many researches are still on the optimization of control algorithm , However, the separation of memory and CPU is the two major bottlenecks in the field of control and artificial intelligence “ Von Neumann ” Low energy efficiency due to storage wall effect caused by architecture , And Moore's law, which leads the development of semiconductors, is expected to fail in the next few years . In view of these two problems , More applicable intelligent control method in the future , obvious , It should be a brain like computing control method based on non von Neumann architecture .
There are three major areas in brain like computing , first , Brain like computing algorithm , chip , Software tool chain and system ; second , Interface ; third , application . Combined with this paper , This paper mainly explains my brain like algorithm , Chip and Application .
One , Brain like computing algorithm
When summarizing various intelligent algorithms , I try to compare the real universe with the digital world , It turns out that the basic principles of the human brain and the computer are exactly the opposite . In the real world , Each floor supports the next , But the digital universe goes back to the bottom . Current artificial neural networks , It's just a description of spatial complexity , Complete a fool like mapping . And for the real brain , One neuron to one thousand ten thousand neurons . This is the concept of expansion and contraction , Computers are shrinking , The brain is enlarged , So the final processing of information is very fast , So we want to build the same architecture , We need to keep the architecture of the computer brain like .
In Physics , This is the increase and decrease of entropy , Reflected on the computer , That's what happened , energy consumption , Speed and bandwidth, etc .
Because of the generalized universal system , There are too many elements to control , Now our cognitive abilities are limited and vary , So according to the traditional intelligent algorithm , The conclusions are mixed . So I want to improve accuracy and cognitive ability , The complexity of time and space must be combined , And the most effective way , Not blindly pursuing theoretical accuracy , It's trial and error , Trial and error is a very important human ability . Therefore, training an algorithm model mainly uses the basic intelligence such as perception and cognition through environmental stimulation and interactive training , It's more efficient , The acquired intelligence is also more adaptable to complex environments .
Combine the complexity of space and time , That is to combine the coding of space with that of time , It is to transform the information of a multidimensional space into 0101 The amount of information in such a , The main frequency is getting faster and faster , This is the multidimensional nature of coding algorithm .
Two , Brain like computing chip
The basic components of brain like computer are neural morphological devices that simulate biological neurons and synapses , Its characteristic size is equal to or smaller than its biological counterpart , It's not only smaller than transistors , Moreover, some device errors do not affect the basic function of the system .
What I have learned is that there are outstanding achievements in brain chip research in China , Brain like intelligent chips of Cambrian and tianjixin of brain like computing center of Tsinghua University . While I'm looking at the basic information of these chips , come to know , Several technologies in the background of brain like intelligent chip development , This is also the core technology for the development of smart chips in the next few years .
1. Development of advanced measurement and calculation , The understanding of the brain is becoming clearer and clearer , It seems to have reached a turning point , It's how humans really understand the basic mechanisms of the brain .
2. The development of supercomputer , You can do a lot of simulation , Save time , Labor saving , save time .
3. Cloud computing and the Internet provide a sufficiently complex system , Deliver to a person , interaction , study , At the same time, its development gives us a lot of enlightenment , We have a good understanding of it .
4. Development of nanodevices , It has been able to make devices with the same level of energy consumption as the human brain .
It is also the technology of all walks of life , Rise together , Let me realize the development of technology , Just like human development , It's not going to be on your own , It's integrated , For brain like intelligence , It has to be a precision instrument , Medical Science , computer , automation , microelectronics , The knowledge of electronics and materials permeates each other , They are compounded with each other .
Three , Application of brain like computing
Always look at the world from the perspective of students , There are many views on education , In fact, only students can deeply understand the problems faced by education , such as , Unfair educational resources , Don't teach students according to their aptitude , Theory is not linked with practice and so on . I've learned a lot of intelligent algorithms , Is there a way to subvert the current education system , How about being educated according to the real needs of our students ? I hope brain like computing can , Not only can it be regarded as the future era of cognitive intelligence , The main means of intelligent association of all things , It can also promote the reform of education methods , Improving technology with thinking , Change thinking with technology .
of course , In the field of brain like computing, self-developed software tool chain and brain like computing system are also very important . Among them, the intelligent algorithm involved at present , Is it just a general artificial intelligence algorithm , Software development is also based on Tensorflow Architectural . therefore , We still have a long way to go in the research of brain like computing model , It is also an urgent need to solve the problem independently , I hope I can make a breakthrough in this direction in the future .