Be 1 300 million times per second, the speed of change in cellular automata cell cavity color; it can handle nearly 100 million personneuron; its computing power equivalent to 10,000 personal computers, it is definitely worth spending a 500 000 dollar supercomputers.
The internal structure of CBM
CBM has two main tasks. The first is to evolve a number of
Individual neural circuit module or modules that I said. A neural network is owned 24 24 24 cells or small cubes of the threedimensional cellular automata cavity growth or evolution. This space can accommodate 1 000 neurons. The same side as dendritic or axonal growth in this space randomly. A programmed FPGA is used to calculate the growth of neural
Network signal quality, the basic idea and I have the work done before 1996 is similar. When a module evolution is complete, it is downloaded to a more than 10 billion bytes of memory. 64 000 such modules will be evolved, one at a time, each has its own part of the human "evolutionary engineer" (Evolutionary Engineers, EE) designated fitness
Belgian laboratory, the fourth hardware platform developed by my colleagues have. Because Belgium has the world at that time four CBM in two, a sense that Belgium is a world leader in this field. In 2000, I successfully received from the Brussels government 100 million in research funding, used to make artificial brain to control a robot like a kitten
The same, it has hundreds of capacity. All of these 20th century work of all my 90’s 80’s work of some of the extension of Dr..
"CBM" brain manufacturing machines, Japan, 1999
CBM in the end be able to do? At that time I believe it is a truly wonderful machine, when people realize its importance, will occupy the important
Position in the history of computing. Electronic technology to achieve it directly on cellbased CoDi neural network model for automated machines. It evolved in a few seconds into a neural network, that is, it can at this time to conduct a complete genetic algorithm operation, that is, tens of thousands of neural circuit module generation and adaptive computing. It can
Work and create FPGA chip U.S. company’s decision to the board removed from the market. My American colleagues and the company had to fight for the remaining number of chips. This delay for several months. The chips eventually get, but have not been tested. Therefore, he had himself tested, because no company’s detailed testing, leading to more delay.
Until mid2000,
I called CAMbrain machines (CBMs) are adequately tested, the real evolution experiment can begin. CAM the Cellular Automata Machines (on cell automata), early artificial brain is on cell automata to achieve.
The first CBM in 1999 was sent to my laboratory in Japan, it still has flaws. Only rarely did not test the chip, and human resources, the slow progress.
But these are not frustrating, because some other people interested in starting on the CBM. In 2001, the world has four such machines. The first stay in my original laboratory at Kyoto, Japan. Second is a speech processing laboratory Belgium to buy, then transferred to a bioinformatics company, is also in the Belgian. Taiwan is my third purchase of the
Academic I am, I refused and told him that the West is considered a very disgusting, an abuse of rights and . After the sharp deterioration in our relationship. I was allowed to stay until the end of 1999, and then I will have to leave. Japan’s economic performance in the 20th century, 90 is so bad, is known as
A "lost decade", so that the research was considered to be bleak, hopeless, and therefore do not need to be granted the period of economic malaise. Therefore, I and departments where most people left the Japanese in the new millennium. Then I got another job, again in Brussels, and is doing the same work and Japanese laboratories.
1996 to 2000,
I colleagues in the United States is responsible for hardware manufacturers to build artificial brains to realize my ideal of special hardware, has been tirelessly working with. For him, progress is slow. His boss from Japan where I received only limited funding, he only requested can afford a fulltime assistant and a few parttime assistant.
In the course of his
Second half of 1996, an American electrical engineer and me. He thinks that my thesis is very interesting and would like to cooperate with me. I put some details of the new model sent to him, and asked whether he was on the market with a particular FPGA board implement it in hardware, he said, should be able to. My
Boss agreed to the idea of Japan and give subsidy. The United States started with my new colleagues closely Unfortunately, a year after the loss of my German counterparts. He returned to Europe for a doctorate. My boss returned to Japan, "a person responsible for a project," the old way, I become more depressed than before the.
I started thinking
About leaving, but it does not, as has just approved making the new machine. I and my team manager relations are increasingly strained, especially when I discovered that he had such a policy, that is, his name added to his subordinates and he wrote some contributions are not academic journal articles. He told me to add his name to an
Everyone is responsible for a project, it makes me feel very lonely and intellectual emptiness. I do not have a really talk to. Put in some pressure on them, I finally got in 1996 a Master’s assistant a young German.
I explained to my assistant, threedimensional version has been basically finished, and I have been using the cellular automata model
Cavity has become increasingly clear. I briefed him on the circuit I would like to direct the generation and evolution of the electronic speed automatic machines based on cell dreamed of neural networks. I think we should simplify the cellular automaton model cavity, it can be adapted to the then new idea of electronic technology in 1996. He wishes to
Listen to my series, and then disappears for two weeks. When he came back, brought a new, more simplified neural network model, it kept my old model where most of the features, but added some new features, making it can be directly realized in electronic devices . This new model is called "CoDi".
Probably during this period, that is, the
To do so, this is achievable.
I spent a year to write all these rules (North East South West the types of rules) to prove the twodimensional version of the automatic machine based on cell neural network is run, can be evolved. I had to handwrite (with software tools to help me) about 11 000 such an order to make
It work, the results it work. I have successfully evolved oscillator circuit, the signal strength detection circuit, linear motion detection circuit, and so on. Is the time to study has a different version of a three dimensional topology. In the twodimensional structure, the circuit will encounter each other, they can not cross another circuit. However, in the threedimensional structure, the
Cellular automata cavity path cross each other through the third dimension. Threedimensional circuits and evolutionary dynamics of twodimensional circuit to be rich than many. Just two years later I asked to work the threedimensional version, this version has about 60 000 orders.
However, this time I was in Japan live very depressed. Our group’s immediate supervisor has a policy that
The fitness value of the entire network, while the network is in turn generate growth in a series of instructions, that is, contains six (all digital) random string of integers.
What I did is to make neural networks and cellular automata combined cavity. These far as I know no predecessors did. Did so because I think that the cellular automaton
Is a cavity can have billions of cells on cell automata suitable media, to the billions of neurons are sufficient. This to me is achievable. At that time the workstation (PC powerful than some of the computer) about one billion bytes of memory. Memory is very cheap, so if I can in a byte (8 bits) is stored on cell
Cellular automata state or color, and my workstation can have gigabytes of memory, then I will be able to store 10 100 million cellular automata cell cavity color, 1 billion! This is a lot of, enough to put in which a man has a lot of brain neurons. Space will not be a problem. At that time the technology allows
Of the growth path, it can extend the path forward, turn left, turn right, split, etc., these decisions on the color of the growth signal. Automatic machine by changing the path on cell growth signals in the middle of the order, I was able to evolve based on cell automata neural network.
This process has two steps. The first is
The growth stage. After a few hundred units of time, growth reached saturation. No three cells wide path on cell growth of the automatic machine. The path is the axonal and dendritic neural network. When the growth phase is complete, that control cell growth and is cleared from the network, the growth of neural networks can be followed by the
Signaling stage to use. Input signal will be imposed, and would spread to the entire network. I like the early years of the evolution of these signals out of the same neural network signal. They can be extracted in the output point, and can be used to control those fitness (fitness) or performance can be measured process. The fitness is
1993, that such plans often provoke distrust of vision. Since then, most of the neural network researchers use only dozens or hundreds of neurons. As I, like a few years ago, when I heard someone say to use billions of neurons, is simply ridiculous. To do this I was laughed at, ridiculed.
However, I firmly believe that my idea is
Feasible. If someone can create an evolution based on the principle of special computer hardware, then the speed of evolution of their electronic nature of manufacturing will make it possible for the brain. I carried out with mathematical reasoning, that by the year 2001 to produce a billion artificial neurons, the brain is feasible. I have a Japanese laboratory 7
To 8 years of the contract, so I think I have enough time to realize my dream.
My first task is to choose a design and evolution of neural networks can be a medium. I chose to use on cell automata (Cellular Automata, or referred to as CA). In the twodimensional cellular automata cavity every cell on the board like