Article from the old book - "How to create an optical brain"
Good evening everyone!
Recently a book came into my hands, an article from which I was so intrigued that I hurried to share it with the community. The book is old - if I'm not mistaken in 1986 release. Of course - this article is a typical scientific speculation - the publication of a poorly proven hypothesis to attract attention, but it intrigued me.
Publish directly from the scanner - one to one. Therefore, a lot of letters.
HOW TO CREATE AN OPTICAL BRAIN "V.M. ZAKHARCHENKO, G.V. SKROTSKY
The successes of neurophysiology in recent years have largely clarified the principles of the brain - the most complex and mysterious of natural phenomena known to us. According to the famous American scientist D. Hubel: in the past decade, neuroscience has become one of the most active branches of science. The consequence of this was a genuine explosion of discoveries and insights. ”
On the other hand, the 70s were characterized by the rapid development of microelectronics, optoelectronics, and optical information processing techniques. Therefore, the attempts to use the achievements of modern technology and technology to model the brain and create on this basis fundamentally new information processing systems are natural and logical. Thus, the combination of the possibilities of optoelectronics and some methods of optical information processing allowed to propose and substantiate a new idea - the idea of ​​creating an optical brain.
As you know, the brain consists of nerve cells - neurons, interconnected by processes of neurons and interneuron connections - synapses. According to the latest data, there are at least 5 • 10 ^ 14 neurons in the brain. Despite their huge number, the bodies of neurons occupy only a few percent of the total brain. The rest of the space is occupied by interneuron connections — nerve fibers of micron and submicron thickness. Each neuron of the cerebral cortex has up to several tens of thousands of connections through which signals come from other neurons. If the total effect of these signals exceeds the threshold of the neuron, then it is excited and generates an output signal. The neuron has only one exit, but it forks into many connections to other neurons. The transmission coefficients of the communication signal are not the same (both in meaning and in sign), therefore, completely different signals arrive at other neurons. Neurons can be compared with control centers that receive and distribute signals coming via inter-neural connections. There are at least 10 ^ 14 such connections in the brain. Understanding that synapses belong to the main structural components of the brain, primarily determining its functional characteristics, is one of the most significant findings made by neurophysiologists. In confirmation, we can cite the statement of the well-known neurophysiologist E. Kandel: “Many neurobiologists are convinced that ultimately it will be proved that the unique properties of each person - the ability to feel, think, learn and remember - are in strictly organized networks of synthetic interconnections between neurons of the brain.”
Most of the brain, approximately 1000 cm ^ 3 out of 1400 cm ^ 3, is occupied by the cerebral cortex. It is folded and has a thickness of about 3 mm. The entire area of ​​the cortex is divided into functional information processing zones: visual, auditory, motor, etc. In turn, the functional zones are divided into modules with an area in fractions of a square millimeter and a height equal to the thickness of the cortex. Each module is responsible for processing a certain type of signals from certain receptors, for example, a portion of the retina.
A huge variety of information about the properties of the environment coming to the brain from the sense organs is displayed on the set of neurons of the cerebral cortex. Depending on the parameters of the incoming signal and on its position in space, certain areas of the cortex are excited. The organization of the cortex vertically layered. Each neuron of one layer is connected mainly with neurons of another layer. An ensemble of excited neurons of one layer sends signals to another layer, an ensemble of excited neurons also appears in the second layer, etc. Each module of the cortex is a local neural network that transforms information by passing it from input to output.
In such a simplified to the limit model of the brain, the problem of developing its artificial analogue in technical terms can be divided into two parts: the creation of artificial neurons and the realization of a spatial structure of tens and hundreds of trillions of neuronal connections.
Various electronic models of neurons have been developed. With the help of modern integrated technology, you can always make them in sufficient quantities. Recreating the spatial structure of neuron connections is an incomparably more difficult task. In the rich arsenal of microelectronic circuit technology, there are no methods that allow systems to be created, each element of which would have thousands and tens of thousands of links with other elements of the system. And not just a connection, but such, each of which has its own individual conductivity. To implement the most complex spatial structure of a huge number of intertwining relationships, fundamentally new solutions are needed.
The real practical way to solve this problem lies in the optical modeling of neural structures. Light rays do not interact with each other, and therefore restrictions on the saturation density of space by optical communication channels and the geometry of their location are completely removed. Techniques created during the development of holographic memory can be used for such modeling. With more or less changes, almost all the numerous variants of holographic memory that exist today can be used. For example, the first experimental model of the neural network was based on the most common scheme of a holographic storage device with a gas-discharge laser, a beam deflector and a rectangular hologram matrix. The most promising technology for creating optical models of neural systems is a technique that uses the capabilities of an integrated technology of micro- and optoelectronics. Therefore, we consider as an example an optical neural network with a holographic memory based on matrices of semiconductor lasers.
Information in such a memory is recorded on a photosensitive medium in holograms (up to 1 mm in diameter) collected in matrices. In front of the hologram matrix there is a matrix of semiconductor lasers. A laser beam, passing through a hologram, splits into a set of light rays, the location and intensity of which depend on the information recorded on the hologram. Behind the hologram matrix at some distance there is a matrix of photocells recording the light signals.
Now imagine that each laser is the output of a particular neuron. Its output signal - the beam - is split in the hologram into a multitude of light-beam connections going to the neuron inputs of the next layer - photocells. Light links are different in their weight-intensity beam. All light signals going to a specific neuron are summed by a photocell, the output of which is proportional to this total signal at the input. So, the input of a neuron is a photocell, and the output is a laser plus a hologram with a fan of the connections of this neuron recorded on it with all the neurons of the next layer. It remains to connect the input with the output, placing a threshold element between them, and we will get a neuron model.
Place another one behind the matrix of the photocells of the holographic memory so that the signals of the photocells of the first memory control the emission of the array of semiconductor lasers of the second. We will place a third memory behind it, and so on. As a result, we will obtain a periodic structure equivalent to a sequence of neural layers of the brain. Here, just as in the brain, the information received at the input is transferred from layer to layer, going through higher and higher processing steps, the program of which is determined only by the structure of the links recorded on the holograms. The density of these bonds is equal to the density of recording information on holograms and is about 104 bonds per 1 mm2.
To change the system of connections, it is enough to replace the hologram block with another. Even the brain created by nature does not have such a technical advantage. True, he has another advantage. All interneuronal connections of the brain are flexible, they can change in the process of training a person, accumulation of life experience. The optical brain is pre-trained, all its knowledge is enclosed in interchangeable blocks of holograms, in the structures of synthetic interconnections recorded on them. If you set the task of creating a complete analogue of the human brain, then such differences, of course, are a drawback. If we keep in mind the technical applications of artificial neural systems, for example, in robotics, where the possibility of mass production and a quick change of the program of behavior of the robot is required, then these differences become an advantage.
The described system has another advantage - modular construction, and the module is a block of holographic memory. Consider the possible parameters of such a module. The recording density of ties on holograms can reach 10,000 links per square centimeter. This means that 1000 holograms can be recorded on a 1 cm2 plate, each of which has 1000 connections connecting 1000 outputs of neurons of one layer with 1000 inputs of neurons of the next layer. The possibilities of modern technology allow to produce a matrix of 1000 lasers on an area of ​​1 cm2. A matrix of 10 photocells on an area of ​​1 cm2 for modern integrated technology has already passed a stage. The task is facilitated by the fact that neither the laser arrays, nor the photocell arrays have external electrical connections, except, of course, power.
Consequently, the module under consideration, let's call it opto-neuron, is equivalent to a layer of a thousand neurons with a million interneuron connections, contains a matrix of thousands of semiconductor lasers, a matrix of thousands of photocells and looks like a cube with a side of 1 cm. The response time of module elements is not more than 10 ^ -6 c, and the number of its elements approximately corresponds to the number of neurons in one layer of the neural module of the cerebral cortex. From modules, like cubes, you can add complex neural structures.
Let us try to estimate the size of an opto-neuron brain model containing 5 • 10 ^ 10 neurons and 5 • 10 ^ 13 interneuron connections. To build such an optical brain, you need 5 * 10 ^ 7 modules for a thousand neurons with a total volume of 50 m2. The volume of modern computers with all the equipment is about the same. Of course, compared with the human brain, which has a volume of about 1.5 liters., The optical brain loses approximately 30,000 times. But we must not forget that in speed of elements, and consequently, in computing power, it is 10 ^ 4 - 10 ^ 5 times higher than the human brain.
Consider now another problem. It is not enough to make an optical brain. In order to breathe life into it, it is necessary to fill it with informational content - to determine the structure of light connections. Then the optical brain will come to life and will do the work that is determined by the nature of its connections: translate from Russian into English, or control a spacecraft, or analyze visual images, etc. But it is much more difficult to determine the structure of brain connections than to create an artificial brain itself. There is a direct analogy with computer technology: the cost of computer software is several times higher than the cost of the machine itself.
There are two main ways to create information processing algorithms in artificial neural systems. The first requires the study of the principles and schemes of information processing in the brain by the methods of neurophysiology. This kind of work is actively underway. An example is the study of the principles of processing in the brain of visual information, carried out by American scientists D. Hubel and T. Wiesel, who were awarded the Nobel Prize in Medicine of 1981. Another way is the analytical conclusion of algorithms that simulate individual brain functions. The simplest of such algorithms include, for example, information retrieval algorithms by keywords used in most of the existing information retrieval systems.
Let us consider a variant of the algorithm developed and practically implemented in the opto-neuron recognition system of search images of documents. Despite its simplicity, this algorithm is in many ways reminiscent of the intellectual operations performed by a person when searching for information.
Imagine that you are looking for literature in a library catalog for a specific query, for example: “Designing Transistor Radio Receivers”. See how you work. First of all, you, of course, read the text of the request and in the process of reading turn the sequences of letters into words denoting concepts. This is the first stage of information processing. Then you remember words that are close in meaning. As a result, not only the words of the request are fixed in your mind, but also many other words and concepts associated with them. For example, if, looking through a catalog, you meet a card for the book “Development of portable radio equipment”, then although this card contains words other than in the request, you will put it off because the word “development” is associated with “design”, the word “portable "Means most likely that the equipment is" transistor ", etc. So, enrichment of the request is the second stage of information processing, which uses your knowledge of the topic. And finally, the third stage of processing is an assessment of the semantic proximity of the contents of the catalog and query cards.
We analyzed the process of searching for information by man and singled out three main stages in it. Now let's try on the basis of this analysis to develop the structure of the neural information retrieval system. Let's write the information processing scheme: letters - words - an associated set of words - cards with addresses of books. Four forms of information and three stages of processing in the transition from one form to another. In neural systems, information is transformed during the transition from layer to layer. This means that our opto-neuron system must contain four neural layers and three hologram matrices with interneuron connections, which fill three interlayer spaces.
The first layer is letters. Each neuron of the first layer corresponds to a specific letter of the alphabet (taking into account its place in the word). The second layer is words. Each neuron of the second layer corresponds to a specific word from the used dictionary. The third layer is also words. And finally, the fourth layer - these are the objects of search - catalog cards. Each neuron of the fourth layer corresponds to a specific catalog card.
Now consider the interneuron connections. First, the connection of the first and second layers. The neuron of the second layer is connected with the neuron of the first layer, if the corresponding letter is included in the corresponding word. Optical communication of the second and third layers - mapping associative links between words in the human brain. If there is an associative connection between two words, then the corresponding neuron of the second layer is connected by a light link of a certain intensity with the corresponding neuron of the third layer. The third set of links between words and search objects reflects the set of keywords contained in the cards. If there is a keyword in the card, then its neuron in the third layer is connected by a light beam to the neuron of this card in the fourth layer.
By writing down holograms with interneuron connections, we thereby entered into the memory of the system the necessary information. Now consider how it works. When you enter the letters that make up the words of the query, the corresponding neurons of the first layer are excited. In this case, the lasers standing at the output of these neurons are turned on. Holograms split the radiation of lasers into a multitude of rays that reach the inputs of the neurons of the second layer in accordance with the interneuron scheme. Neurons are excited in it, to the inputs of which a total signal has been received that exceeds the threshold of neurons. The ensemble of excited neurons of the second layer corresponds to the set of words of the query. The light connections of the neurons of this layer fall on the neurons of the third layer and also excite some of them. The ensemble of excited neurons of the third layer corresponds to the associated set of words, and the ensemble of excited neurons of the fourth layer corresponds to catalog cards that meet the query. The lasers included at the output of the neurons of this layer denote the cards found.
Compare the capabilities of modern computing, the human brain and the optical brain. Comparison will be carried out in two key parameters: the speed of information processing and memory. For computers using a digital information processing mechanism, these parameters are determined by the number of arithmetic operations per second and the amount of memory in bits. For the brain working on other principles, these parameters are not defined. , , , — , , . , . , , . .
. . . , — . , , . . . , 10^14, — 100 ^-1, 10^16 .
, , . 10^14 50 , 10^16 . , 10^6 -1,: 10^20 . 10^9 10^9 10^20 , . , , , , .
. ? , — . , , .
Recently a book came into my hands, an article from which I was so intrigued that I hurried to share it with the community. The book is old - if I'm not mistaken in 1986 release. Of course - this article is a typical scientific speculation - the publication of a poorly proven hypothesis to attract attention, but it intrigued me.
Publish directly from the scanner - one to one. Therefore, a lot of letters.
HOW TO CREATE AN OPTICAL BRAIN "V.M. ZAKHARCHENKO, G.V. SKROTSKY
The successes of neurophysiology in recent years have largely clarified the principles of the brain - the most complex and mysterious of natural phenomena known to us. According to the famous American scientist D. Hubel: in the past decade, neuroscience has become one of the most active branches of science. The consequence of this was a genuine explosion of discoveries and insights. ”
On the other hand, the 70s were characterized by the rapid development of microelectronics, optoelectronics, and optical information processing techniques. Therefore, the attempts to use the achievements of modern technology and technology to model the brain and create on this basis fundamentally new information processing systems are natural and logical. Thus, the combination of the possibilities of optoelectronics and some methods of optical information processing allowed to propose and substantiate a new idea - the idea of ​​creating an optical brain.
As you know, the brain consists of nerve cells - neurons, interconnected by processes of neurons and interneuron connections - synapses. According to the latest data, there are at least 5 • 10 ^ 14 neurons in the brain. Despite their huge number, the bodies of neurons occupy only a few percent of the total brain. The rest of the space is occupied by interneuron connections — nerve fibers of micron and submicron thickness. Each neuron of the cerebral cortex has up to several tens of thousands of connections through which signals come from other neurons. If the total effect of these signals exceeds the threshold of the neuron, then it is excited and generates an output signal. The neuron has only one exit, but it forks into many connections to other neurons. The transmission coefficients of the communication signal are not the same (both in meaning and in sign), therefore, completely different signals arrive at other neurons. Neurons can be compared with control centers that receive and distribute signals coming via inter-neural connections. There are at least 10 ^ 14 such connections in the brain. Understanding that synapses belong to the main structural components of the brain, primarily determining its functional characteristics, is one of the most significant findings made by neurophysiologists. In confirmation, we can cite the statement of the well-known neurophysiologist E. Kandel: “Many neurobiologists are convinced that ultimately it will be proved that the unique properties of each person - the ability to feel, think, learn and remember - are in strictly organized networks of synthetic interconnections between neurons of the brain.”
Most of the brain, approximately 1000 cm ^ 3 out of 1400 cm ^ 3, is occupied by the cerebral cortex. It is folded and has a thickness of about 3 mm. The entire area of ​​the cortex is divided into functional information processing zones: visual, auditory, motor, etc. In turn, the functional zones are divided into modules with an area in fractions of a square millimeter and a height equal to the thickness of the cortex. Each module is responsible for processing a certain type of signals from certain receptors, for example, a portion of the retina.
A huge variety of information about the properties of the environment coming to the brain from the sense organs is displayed on the set of neurons of the cerebral cortex. Depending on the parameters of the incoming signal and on its position in space, certain areas of the cortex are excited. The organization of the cortex vertically layered. Each neuron of one layer is connected mainly with neurons of another layer. An ensemble of excited neurons of one layer sends signals to another layer, an ensemble of excited neurons also appears in the second layer, etc. Each module of the cortex is a local neural network that transforms information by passing it from input to output.
In such a simplified to the limit model of the brain, the problem of developing its artificial analogue in technical terms can be divided into two parts: the creation of artificial neurons and the realization of a spatial structure of tens and hundreds of trillions of neuronal connections.
Various electronic models of neurons have been developed. With the help of modern integrated technology, you can always make them in sufficient quantities. Recreating the spatial structure of neuron connections is an incomparably more difficult task. In the rich arsenal of microelectronic circuit technology, there are no methods that allow systems to be created, each element of which would have thousands and tens of thousands of links with other elements of the system. And not just a connection, but such, each of which has its own individual conductivity. To implement the most complex spatial structure of a huge number of intertwining relationships, fundamentally new solutions are needed.
The real practical way to solve this problem lies in the optical modeling of neural structures. Light rays do not interact with each other, and therefore restrictions on the saturation density of space by optical communication channels and the geometry of their location are completely removed. Techniques created during the development of holographic memory can be used for such modeling. With more or less changes, almost all the numerous variants of holographic memory that exist today can be used. For example, the first experimental model of the neural network was based on the most common scheme of a holographic storage device with a gas-discharge laser, a beam deflector and a rectangular hologram matrix. The most promising technology for creating optical models of neural systems is a technique that uses the capabilities of an integrated technology of micro- and optoelectronics. Therefore, we consider as an example an optical neural network with a holographic memory based on matrices of semiconductor lasers.
Information in such a memory is recorded on a photosensitive medium in holograms (up to 1 mm in diameter) collected in matrices. In front of the hologram matrix there is a matrix of semiconductor lasers. A laser beam, passing through a hologram, splits into a set of light rays, the location and intensity of which depend on the information recorded on the hologram. Behind the hologram matrix at some distance there is a matrix of photocells recording the light signals.
Now imagine that each laser is the output of a particular neuron. Its output signal - the beam - is split in the hologram into a multitude of light-beam connections going to the neuron inputs of the next layer - photocells. Light links are different in their weight-intensity beam. All light signals going to a specific neuron are summed by a photocell, the output of which is proportional to this total signal at the input. So, the input of a neuron is a photocell, and the output is a laser plus a hologram with a fan of the connections of this neuron recorded on it with all the neurons of the next layer. It remains to connect the input with the output, placing a threshold element between them, and we will get a neuron model.
Place another one behind the matrix of the photocells of the holographic memory so that the signals of the photocells of the first memory control the emission of the array of semiconductor lasers of the second. We will place a third memory behind it, and so on. As a result, we will obtain a periodic structure equivalent to a sequence of neural layers of the brain. Here, just as in the brain, the information received at the input is transferred from layer to layer, going through higher and higher processing steps, the program of which is determined only by the structure of the links recorded on the holograms. The density of these bonds is equal to the density of recording information on holograms and is about 104 bonds per 1 mm2.
To change the system of connections, it is enough to replace the hologram block with another. Even the brain created by nature does not have such a technical advantage. True, he has another advantage. All interneuronal connections of the brain are flexible, they can change in the process of training a person, accumulation of life experience. The optical brain is pre-trained, all its knowledge is enclosed in interchangeable blocks of holograms, in the structures of synthetic interconnections recorded on them. If you set the task of creating a complete analogue of the human brain, then such differences, of course, are a drawback. If we keep in mind the technical applications of artificial neural systems, for example, in robotics, where the possibility of mass production and a quick change of the program of behavior of the robot is required, then these differences become an advantage.
The described system has another advantage - modular construction, and the module is a block of holographic memory. Consider the possible parameters of such a module. The recording density of ties on holograms can reach 10,000 links per square centimeter. This means that 1000 holograms can be recorded on a 1 cm2 plate, each of which has 1000 connections connecting 1000 outputs of neurons of one layer with 1000 inputs of neurons of the next layer. The possibilities of modern technology allow to produce a matrix of 1000 lasers on an area of ​​1 cm2. A matrix of 10 photocells on an area of ​​1 cm2 for modern integrated technology has already passed a stage. The task is facilitated by the fact that neither the laser arrays, nor the photocell arrays have external electrical connections, except, of course, power.
Consequently, the module under consideration, let's call it opto-neuron, is equivalent to a layer of a thousand neurons with a million interneuron connections, contains a matrix of thousands of semiconductor lasers, a matrix of thousands of photocells and looks like a cube with a side of 1 cm. The response time of module elements is not more than 10 ^ -6 c, and the number of its elements approximately corresponds to the number of neurons in one layer of the neural module of the cerebral cortex. From modules, like cubes, you can add complex neural structures.
Let us try to estimate the size of an opto-neuron brain model containing 5 • 10 ^ 10 neurons and 5 • 10 ^ 13 interneuron connections. To build such an optical brain, you need 5 * 10 ^ 7 modules for a thousand neurons with a total volume of 50 m2. The volume of modern computers with all the equipment is about the same. Of course, compared with the human brain, which has a volume of about 1.5 liters., The optical brain loses approximately 30,000 times. But we must not forget that in speed of elements, and consequently, in computing power, it is 10 ^ 4 - 10 ^ 5 times higher than the human brain.
Consider now another problem. It is not enough to make an optical brain. In order to breathe life into it, it is necessary to fill it with informational content - to determine the structure of light connections. Then the optical brain will come to life and will do the work that is determined by the nature of its connections: translate from Russian into English, or control a spacecraft, or analyze visual images, etc. But it is much more difficult to determine the structure of brain connections than to create an artificial brain itself. There is a direct analogy with computer technology: the cost of computer software is several times higher than the cost of the machine itself.
There are two main ways to create information processing algorithms in artificial neural systems. The first requires the study of the principles and schemes of information processing in the brain by the methods of neurophysiology. This kind of work is actively underway. An example is the study of the principles of processing in the brain of visual information, carried out by American scientists D. Hubel and T. Wiesel, who were awarded the Nobel Prize in Medicine of 1981. Another way is the analytical conclusion of algorithms that simulate individual brain functions. The simplest of such algorithms include, for example, information retrieval algorithms by keywords used in most of the existing information retrieval systems.
Let us consider a variant of the algorithm developed and practically implemented in the opto-neuron recognition system of search images of documents. Despite its simplicity, this algorithm is in many ways reminiscent of the intellectual operations performed by a person when searching for information.
Imagine that you are looking for literature in a library catalog for a specific query, for example: “Designing Transistor Radio Receivers”. See how you work. First of all, you, of course, read the text of the request and in the process of reading turn the sequences of letters into words denoting concepts. This is the first stage of information processing. Then you remember words that are close in meaning. As a result, not only the words of the request are fixed in your mind, but also many other words and concepts associated with them. For example, if, looking through a catalog, you meet a card for the book “Development of portable radio equipment”, then although this card contains words other than in the request, you will put it off because the word “development” is associated with “design”, the word “portable "Means most likely that the equipment is" transistor ", etc. So, enrichment of the request is the second stage of information processing, which uses your knowledge of the topic. And finally, the third stage of processing is an assessment of the semantic proximity of the contents of the catalog and query cards.
We analyzed the process of searching for information by man and singled out three main stages in it. Now let's try on the basis of this analysis to develop the structure of the neural information retrieval system. Let's write the information processing scheme: letters - words - an associated set of words - cards with addresses of books. Four forms of information and three stages of processing in the transition from one form to another. In neural systems, information is transformed during the transition from layer to layer. This means that our opto-neuron system must contain four neural layers and three hologram matrices with interneuron connections, which fill three interlayer spaces.
The first layer is letters. Each neuron of the first layer corresponds to a specific letter of the alphabet (taking into account its place in the word). The second layer is words. Each neuron of the second layer corresponds to a specific word from the used dictionary. The third layer is also words. And finally, the fourth layer - these are the objects of search - catalog cards. Each neuron of the fourth layer corresponds to a specific catalog card.
Now consider the interneuron connections. First, the connection of the first and second layers. The neuron of the second layer is connected with the neuron of the first layer, if the corresponding letter is included in the corresponding word. Optical communication of the second and third layers - mapping associative links between words in the human brain. If there is an associative connection between two words, then the corresponding neuron of the second layer is connected by a light link of a certain intensity with the corresponding neuron of the third layer. The third set of links between words and search objects reflects the set of keywords contained in the cards. If there is a keyword in the card, then its neuron in the third layer is connected by a light beam to the neuron of this card in the fourth layer.
By writing down holograms with interneuron connections, we thereby entered into the memory of the system the necessary information. Now consider how it works. When you enter the letters that make up the words of the query, the corresponding neurons of the first layer are excited. In this case, the lasers standing at the output of these neurons are turned on. Holograms split the radiation of lasers into a multitude of rays that reach the inputs of the neurons of the second layer in accordance with the interneuron scheme. Neurons are excited in it, to the inputs of which a total signal has been received that exceeds the threshold of neurons. The ensemble of excited neurons of the second layer corresponds to the set of words of the query. The light connections of the neurons of this layer fall on the neurons of the third layer and also excite some of them. The ensemble of excited neurons of the third layer corresponds to the associated set of words, and the ensemble of excited neurons of the fourth layer corresponds to catalog cards that meet the query. The lasers included at the output of the neurons of this layer denote the cards found.
Compare the capabilities of modern computing, the human brain and the optical brain. Comparison will be carried out in two key parameters: the speed of information processing and memory. For computers using a digital information processing mechanism, these parameters are determined by the number of arithmetic operations per second and the amount of memory in bits. For the brain working on other principles, these parameters are not defined. , , , — , , . , . , , . .
. . . , — . , , . . . , 10^14, — 100 ^-1, 10^16 .
, , . 10^14 50 , 10^16 . , 10^6 -1,: 10^20 . 10^9 10^9 10^20 , . , , , , .
. ? , — . , , .
')
Source: https://habr.com/ru/post/58720/
All Articles