Chip made brain: science fiction into reality?
The human brain that pride, a powerful and mysterious there has been an obsession with information processing functions.Perhaps because of this obsession, whenever new technologies appear, they will be used to attempt to replicate the brain.While in the process, we gain a lot of new information processing tools, but the "copy" and the real brain is still far, and it has proved that the human's understanding of the brain is just at the stage of enlightenment "two children arguing day".
However, in the last six months, there has been a new "building brain action" progress: two giant IBM and Qualcomm has issued the "chip" based on the technology of neural mimicry.IBM's SyNAPSE chip which simulates the one million neurons and more than two hundred million synapses connections, begun to take the size of the human brain.Qualcomm Zeroth released April also on the silicon chip effectively simulate the brain neurons, loaded with the chip machine car can also use "by the inspiration of the human brain algorithm" for pathfinding, avoiding obstacles and other tasks.If people through decades ago to come over, I'm afraid I would think that the overall life science fiction.
Left: IBM released in August 2014 and SyNAPSE chip appearance;Right: qualcomm Zeroth chip in the promotion of the figure
So, "the brain" time really want to come?Conclusion now seems premature.On the simulation of the human brain, scientists are still faced with many challenges and confusion.Created now, let's see "the brain" the science subjects in the real world after all how the status quo.
"Mimicry" hero spectrum chip
Modelled on the life of the architecture of the nervous system to design the large scale integrated circuit (VLSI) hardware electronic technology, known as
Neuromorphic engineering, the engineering pioneered in the 1980 s.In nearly 40 years of development, nerve mimicry chips have repeatedly, especially in recent 10 years, electronic hardware giants such as IBM, HP, qualcomm, nerve starts to take on the lively scene of the accelerated runaway mimicry field.
Nerve mimicry is the key to manufacture can produce "quasi real neurons" of nerve signal, to achieve this goal, there are two ways to go.The first is the use of silicon semiconductor properties, directly on the silicon components with the accumulation of voltage to simulate neuron membrane potential, this way is called the "analog neural mimicry".This solution is the most classic neural mimicry engineering technical route, which directly converts nerve cells in the way of signal transduction in silicon on the conductor.In simulation mode made "neurons" can easily reach the ame speed and life, even faster. s
Another simulation approach is to create a similar small computer digital chip, and then run on neuron simulation program, the simulation program is responsible for the generation similar impulse signal, this scheme is known as "digital nerve mimicry".First mentioned IBM SyNAPSE and qualcomm Zeroth chips used are digital scheme of mimicry.The advantages of digital simulation can adopt various flexible neural model, such as the demand is higher in the simulated applications can join the synapses and detailed characteristic of ion channels, while higher requirements on speed when can simplify the model to ensure the speed.Digital scheme simulation speed decreased, but can also be achieved after optimization and neurons or faster operation speed, and higher flexibility and let it became a popular option of neural mimicry.
Among various kinds of nerve mimicry chip, the vast majority are hybrid integrated both digital and analog techniques, through the complementary advantages make the chip to achieve better performance.
Made enormous challenge of the brain
Now, there has been a considerable "brain" of the chip, they look very "cool", but never seems to be able to shake the status of the conventional computer chips.This is largely due to "build the brain" this mission brought about by the huge challenge.
For many music enthusiasts, Monster this headset brand should be
familiar.Create the magic sound brand is California Chinese engineers Noel Lee.When lee young small fussy regarding the quality of the music has come to the point where harsh, that are found in ordinary people do not pay attention to the wire can improve the sound quality.So the magic sound companies a star's main product is not the familiar headphones, it is thick and strong high-end wire (Monster Cable).The magic sound companies in technological process has provoked a thought: the high quality of information processing systems often require high quality of the signal path,
transmission channel and sometimes even more than the importance of information processing itself.
Unfortunately, the establishment of pathways in the electronic information system is more troublesome also more expensive.If use electronic access to imitate a N neurons in neural network, then all neurons between two communication requires N2 pathway.If every pathway with a physical connection to set up, then simulate the number of neurons in a slightly more, the line would have made a mess.The semiconductor chip technology is essentially two-dimensional wiring, so on a slice of silicon wafer can allow line resources are limited.If you want to use such a pathway to realize the human brain connectivity, will almost certainly be in the laws of physics.
In order to circumvent the plight of the physical connection, a lot of nerve mimicry chip adopted the "Internet" : first to neurons in the "address", and then use router to distribute information.Although this solution avoids the complexity of the circuit, but it is time to change the nature of space, if you don't want to take the article N2 physical pathway, then take N2 times longer to deal with the routing.As the growth of the simulation of neuron, always around but still square level growth of "dimension disaster".So nervous mimicry chip is often the key to success is not how many neurons can make a, but how to efficiently handle information interaction between neurons.IBM's SyNAPSE chip integrates 256 million synaptic connections, the order of magnitude of information interaction is quite a remarkable achievement.
Road of the brain not only subject to the laws of physics, and there are also many disputes on the evaluation criteria.Team from all over the world in the brain research on the subject was very noisy: some stormed the number of neurons and neural coupling simulation, some focused on molecular dynamics modeling of synapses, some focuses on brain plasticity is the ability to learn...You're welcome to say,
however, many created "brain" project a little tree target themselves mean.Even with all these efforts declare success may be developed in the human brain can only in the researchers themselves delineate rules do demonstration, their function is still has many limitations.
Brain target confusion
In addition to technical difficulties, making the target of artificial brain is a problem worth thinking about.In addition to the purely for research, "artificial brain" can also do something for us?
Today's popular electronic equipment and there is no similar, on the human brain in the form on the calculation principle is also the basic not outsourcing, however these electronic devices in quite a number of tasks has done rather well.These tasks include equipment control, large batch processing, repeated work for a long time, and so on.Both in the automatic production workshop, or operating thousands of network computer "server farms", even in the more trendy unattended large warehouse, the original completed by "the human brain and human flesh" have now been passed on to the task of mechanical and electrical equipment, efficient, accurate and reliable has already reached the height of the to reach the brain.
The mechanization of the task without creativity, and asked the operator must be accurate and not tired, they are born to machine design.By contrast, explore, learn and adapt to the environment is the area of the brain with a real advantage.If the "brain" simulation, when the people or the traditional mechanization task to them, then I am afraid, "simulation" lost its meaning.
Unattended, based on the honeycomb intelligent modern warehouse (Kiva Systems, LLC)
Learn to create: "the human brain model" is not better
However, in the field of learning and creating, imitate "the brain model" may not have more advantages than the existing computer technology.
Sweeps across the whole world with the concept of "big data", brewing for many years in the laboratory of machine learning technology finally out of the lab and into the practical application.Many otherwise known as "the brain dominance" work, such as face recognition, chess, music creation, etc., are now gradually appeared instead of by computer.Computer and accomplish these tasks, also did not take the human brain information processing model.
Why should such as exploration, learning, creating domain will be breached by the human brain to?It dates back to the human brain an underlying defect - memory ability is insufficient.
Brain is the basic component of neurons, the cell can form efficient information network, but it's not a very good memory element.Neurons from the function, such as a mouthpiece, ushered in the head, you become a voice message on the other side in it.This kind of similar to the structure of the "filter" can quickly complete some signal transformation, but is not suitable for long-term storage information.Electronic components, in contrast, or magnetic components can keep for a long time in "high potential" and "low" or "South Pole", "North Pole" status, information once written a long time can't forget.Because of this, the computer was born can do it "never forgets anything," while the human brain to pay a lot of energy to consolidate memories.
In addition, the state of electromagnetic components as long as through simple changes can turn within a few microseconds, so information not only longer, and can quickly make changes.However, in the neurons of the brain network, want to quickly change state is not so easy.An accepted way is by "neural plasticity"to change the activity of neurons, but this mechanism requires the input and output end of the cell pulse coupled happened thousands of times, when these coincidences happen to some influence to the strength of the synapses, time-consuming also vary from several minutes to several decades.So from the deposited in the speed of information, computer born confindent, "and if a man can be so estimated would have the strongest" brain ".
As knowledge face, play chess, composition of these tasks is to learn, lenovo and creation, but more important is to accumulate huge information database as support.Due to the deficiency of long memory and ability of rapid solidification, the brain takes longer than computer many times to reserve enough data.If in the field of simulation of the brain research, solve the problem of insufficient memory ability, so "artificial brain" on these learning tasks would be no advantage.
The human brain has the lead project: motion control
In the function of the human brain, then, what on earth is still beyond the reach of electronic equipment?A shining beacon stands alone on the sea of human brain control of body movements.
The movement of the human body neural signal transmission quality in fact than electronic equipment.Nerve impulses in the body for experiencing the effects of various noise and disturbance, you will find that if you listen for intubation pulse rules of neurons are often not much better than "listening to the west".In addition, the delay neural signals is also very significant, the transmission of nerve impulses speed on average only a few tens of meters/seconds, even finish one of the most simple spinal reflex takes off 30 milliseconds.In this 30 milliseconds, a not too high the embedded controller (assuming 1000 hz sampling rate) has completed the 30 times micro control.
Under the disadvantage of these huge, however, the human unexpectedly also can finish all kinds of running jumper twist knead, action can also according to the change of the external environment in the process of adjusting action.These challenges is mechanical bumps and silicon head's nightmare.
If you really want to use the chip to make simulation "brain", athletic ability is probably a good research direction.And simulation of neural network can make the robot movement, it still needs more research to jump to conclusions.
Ultimate to brain
Brain science is growing by leaps and bounds in recent 40 years, the development of the brain be uncovered the mysterious veil of have a little.Computer science is also a soldier, meanwhile, will be more wide, many traditional concept of "exclusive" human brain function in the human brain also realized in the system.Under the double flank of brain science and computer science, electronic technology is used to "re-invent the brain" this ancient also compare the technology of alternative genre, it need to redefine the goal of, also have to rethink the meaning of.
From the point of view, "quasi real brain" should help humans uncovered the mystery of the brain.And from a practical point of view, the purpose of the brain, from the brain should be the real strong functions, create a new system that goes beyond the human brain and computer intelligent devices.As mentioned earlier, our enthusiasm for brain worship began to cool in front of every engineering breakthroughs, but in the body motion control, the performance of the "model" the brain is still worth looking forward to.
If you want to create a human brain, the performance of the chip, the researchers still have a long way to go.But from another perspective, people seem to have been inadvertently make things run like the human brain - the Internet.In fact, the Internet is more and more like a human brain: the number of nodes on large scale;Any
interconnection between each node and highly collaborative;Network structure constantly adjust itself according to the actual demand;Fully independent concurrent between nodes, coordinate but not belonging to each other.Global estimates the number of computers connected to the Internet in 2012 is 1 billion, although tens of billions of neurons in the brain and gap, but also be certain the size of the "neural network".We're here to chip made brain racking their brains, the whole human perhaps has already become another scale self-organizing "brain".