(Original title: Where does the brain exist for memory?)
Written by Rodrigo Quian Quirogo, tzhak Fried) Christof Koch?
Translation? Lu Huimin?
Reviewer Guo Aike
In the past, Russia had a famous neurosurgeon called Akakhi Akakhievitch. There is a weird patient who hopes that Akanovic will help him completely forget his imperious, nasty mother. Akanivić responded to his request and opened the patient's head, removing thousands of neurons one by one. These neurons were related to the patient's memory of his mother. After the surgery, the patient wakes from anesthesia and a miracle arises. The patient loses all memory about his mother, whether it is good or bad memory. Akanovich was very pleased with the success of the operation. When he was happy, he decided to devote himself to the next study to find out the neurons related to the memory of his grandmother.
This story is of course fictional. In 1969, when neuroscientist Jerry Lettvin (deceased) spoke at the Massachusetts Institute of Technology, he told the story about his later theory of grandmother cells. Leitman believes that each of our daily conscious experiences, thoughts, and memories, whether it is for a relative or friend, or any other person or thing, has only about 18,000 neurons corresponding to it. However, Litevin later did neither further proof nor abandon his bold assumptions. For more than 40 years, scientists have always had different views on the theory of “grand cellâ€.
The idea that neurons store memory in a very specific and unambiguous way can be traced back to the theory of so-called "pontificial cells" proposed by William James in the late 19th century. The theory holds that people's consciousness is generated by the "Pope cells." However, both the “grandmother cell†and the “pope cell†hypothesis were contrary to the prevailing theories at that time, namely, the “billion-dollar neuron democracy†proposed by Nobel laureate Charles Sherrington in 1940. The theory of (a millionfold democracy). This theory holds that the perception of any person or thing depends on the cooperation of billions of neurons. In this case, the activity of any single neuron is meaningless, and only the cooperation of large-scale neuron groups can create meaning.
How does the brain store a specific concept? Is it stored through a small number of neurons (e.g., a few thousand or even fewer neurons) or is it using a large number of neurons (hundreds of millions of neurons) distributed throughout the brain? Neuroscientists have been arguing over this issue. However, this controversy also brings benefits, allowing scientists to have a new understanding of memory and conscious thinking. Interestingly, Hollywood also helped a little while in the process.
The neurons that discharge to the actress
A few years ago, we were with Gabriel Kreiman (now an associate professor at Harvard Medical School in the United States) and Leila Reddy, now a researcher at the Brain and Cognitive Center of Toulouse in France. Collaboration, an unusual experiment was completed, and a very interesting neuron was discovered in the hippocampus of a patient's brain (a brain area associated with memory). This neuron only affects the American actress Jennifer. An image of Jennifer Aniston reacted strongly and was indifferent to pictures of other things (dozens of other male actresses, celebrities, places, or animals). In the other patient's hippocampus, a special neuron was discovered. It was only discharged when the actress Halle Berry's picture appeared, even when Berry's name was displayed on the computer screen. And keep silent on other things. There is also a neuron who only responds to actress Oprah Winfrey when her picture appears, or her name is displayed on a computer screen, or Oprah Winn, who is synthesized by a computer. When Frei's pronunciation, this neuron will be discharged. In addition, scientists have also discovered a neuron that only displays images of Luke Skywalker (the character in Star Wars), or displays his name on a computer screen, and broadcasts sounds synthesized by computers. Discharge. There are many similar examples.
This type of observational study can be achieved by directly recording the discharge of a single neuron. Other more common techniques, such as brain functional imaging techniques, can observe the activity of the entire brain area when the subject performs a specific task. Functional imaging of the brain can track the overall energy consumption of the area of ​​excitement in the brain (usually containing millions of neurons), but it cannot distinguish the activity of a small group of neurons, let alone a single neuron. In order to record the electrical impulses delivered by a single neuron, microelectrodes smaller than the hair need to be implanted in the brain. This technique is not as common as functional imaging of the brain. Microelectrodes are implanted into the patient's brain only during special treatments.
Occasionally, there is an opportunity to treat epilepsy patients. When the patient's epilepsy has a strong seizure and normal treatment fails to control the symptoms, surgery is required. In some cases, removal of epileptogenic lesions is feasible and may even cure the patient. Before surgery, doctors need to accurately locate the origin and location of seizures through various techniques. Of course, doctors would prefer non-invasive techniques such as brain functional imaging to perform pre-operative evaluation tests, comprehensive consideration of various test indicators, and analysis of pathological neuroelectric activity (epilepsy) through EEG recordings of the patient's scalp. At the time of onset, a large number of neurons were simultaneously intensively discharged. However, sometimes, relying on non-invasive techniques is not enough to accurately locate epilepsy lesions, at this time, neurosurgeons can only seek help microelectrodes. They implanted microelectrodes deep in the patient's brain and allowed the patient to stay in the hospital for continuous monitoring of the patient's brain activity. The epilepsy was then analyzed based on monitoring data.
During the patient's stay in hospital, sometimes scientists invite patients as volunteers, participate in research experiments, allow them to perform multiple cognitive tasks, and monitor their brain activity. At the University of California, Los Angeles, we used a unique technique in which very thin metal-guided flexible flexible electrodes were implanted in the brain of volunteers for recording. The technology was invented by Freud. He led the Epilepsy Surgery Program at the University of California, Los Angeles, and collaborated with scientists from around the world, including the Koch research group at the California Institute of Technology, and the UK University of Quinn Quirogog researchers. Using this technique, we were able to directly record the discharge of a single neuron while the brain was performing different tasks. In the experiment, the patient looked at the image displayed on the Laptop screen, recalled or performed other tasks, and we continuously monitored the patient's neuron activity. It was in this study that we discovered "Jennifer Aniston Neurons," and our findings inadvertently rekindled the controversy caused by Litevin's "Grandmother Cell" theory.
Re-understand "grand cell"
Can neurons like “Jennifer Aniston Neurons†be “grandmother cells†that scientists have long argued about? In order to answer this question, we must first give a precise definition of "grand cell." An extreme explanation for the "grand cell" hypothesis is that a neuron corresponds to a concept. But since we can find a single neuron and it only excitement for Jennifer Aniston, then we have reason to conclude that there must be more Jennifer Aniston neurons, because in billions The probability of finding one of the neurons and the only specific neuron is almost zero. In addition, if only one neuron is responsible for handling all the information related to Jennifer Aniston, then if the neuron is damaged due to illness or accident, the entire memory of Jennifer Aniston will not disappear. ,How can this be?
Another less extreme explanation for the "grand cell" hypothesis is that any concept has several neurons corresponding to it. This explanation may be reasonable, but it is hard to prove or even impossible to prove. Because we can't try all the concepts, we prove that one neuron only discharges one concept (such as Jennifer Aniston). In fact, the opposite is true. We often find neurons that can discharge more than one concept. Therefore, if a neuron is found in one experiment to discharge only one person, then we cannot rule out that it may also discharge other stimuli, but we did not use this stimulus in the experiment.
For example, on the second day we found Jennifer Aniston Neurons, we repeated experiments. In this experiment, we used a lot of pictures related to her and found out that “Jennifer Aniston Neurons†will also star in Lisa Kudrow (played with Jennifer Aniston). "Friends", both of whom became famous.) Discharge; The neuron that responded to Luke Skywalker was also to Yoda, the character in the movie "Star Wars", and was similar to Luke Skywalker. A Jedi) discharged; another neuron excited about two basketball players; and another neuron excited about one of the authors of Quinn Quirog and his collaborators, both at the University of California, Los Angeles. The patient who volunteered to participate in the experiment had contact with him. Despite this, people can still think that these neurons are "grandmother cells," but there are more than one object that can make them excited and discharged. For example, in the TV drama "Friends," two blonde actress stars in the movie Star Wars. Jedi warriors, basketball players, or scientists who experiment with patients. Therefore, the question of whether these cells are "grandmother cells" seems to have become a semantic problem whether to extend the definition.
For the moment, let's take a look at semantics. Let's focus on some of the key features of these Jennifer Aniston neurons. First, we have found that the excitation of such neurons is very selective. Each one is excited only by pictures of a small group of socialites, politicians, relatives or landmarks displayed to the patient. Second, each of these neurons can be excited by multiple expressions of a particular person or place, regardless of the specific visual characteristics of the picture. In fact, a neuron can produce similar reactions to various pictures of the same person, even his name (whether it is written or read). It's as if this neuron told us in its discharge mode, "I know Jennifer Aniston, no matter what form you use for display: her picture of her red dress, her silhouette, her name written out of it. Even shouting out her name can be." This kind of neuron seems to be discharging a certain concept—no matter what form the concept is expressed. Therefore, it may be more appropriate to rename these neurons as "concept cells" instead of "grand cells." "Concept cells" are sometimes excited by multiple concepts. In this case, multiple concepts are often closely related.
Concept coding
To understand how a small number of neurons relate to a specific concept (such as Jennifer Aniston), we first need to understand a complex process: In our daily lives, how our brain acquires and stores large numbers of people and things Image information. The information the eye sees first passes through the optic nerve behind the eyeball and into the primary visual cortex located in the hindbrain. The neurons here discharge certain tiny details of the image. Each neuron is like a pixel of a digital image, or a colored dot in the painter George Seurat's.
A single neuron does not tell us whether the details it receives correspond to a face, a cup of tea, the Eiffel Tower, or any other image. However, each neuron's information is part of the overall image. When combined, it produces a beautiful image, such as "A Sunday Afternoon On the Island of La Grande Jatte, George. Pull's representative painting.) If the image changes slightly, some of the details of the image will also change. At this time, the discharge of neurons on the primary visual cortex will change accordingly.
The brain needs to process sensory information to get deeper information than the image—it must recognize the target and integrate it into known concepts. From the primary visual cortex, the neuron activity triggered by the image sequentially passes through a series of areas on the cerebral cortex and spreads to the forehead area of ​​the brain. In these more advanced visual areas, a single neuron discharges the entire face or object rather than local details. In these areas, only one neuron is needed to tell us whether the image is a human face or the Eiffel Tower. If you change the image slightly, such as moving the image, or changing the light, the details of the image will change, but these neurons do not seem to mind slight changes in the details of the image, and their discharge conditions remain almost unchanged. The property is called "visual invariance".
The neurons in the advanced visual area communicate their information to the medial temporal lobe, the hippocampus and its surrounding cortex. These areas are related to memory function, and we found it here. Aniston Neuron." The response of hippocampal neurons is more specific than that of neurons in the advanced visual cortex. Each hippocampal neuron only discharges a specific person or, more precisely, discharges that person’s corresponding “conceptâ€: not only the face, but also all aspects of the appearance, but also includes a close relationship with the person. Various attributes, such as the person's name.
We are trying to figure out, in the brain, what is the sparseness of the neuron that encodes the concept? In other words, how many neurons discharge can represent a particular concept. Obviously, we cannot directly measure the number of such neurons because we cannot record the activity of all neurons in a given brain region. However, the author Koch once used statistical methods together with Stephen Waydo (who was a Ph.D. student at Caltech) to estimate that in the medial temporal lobe, a specific concept would trigger less than 1 million The neurons are discharged, and there are about 1 billion neurons in this area. Moreover, considering that the picture used by the researchers in the experiment is very familiar to the patient, which tends to cause more neurons to be discharged, the “1 million†should be an upper limit, which actually represents the number of neurons that define a concept. It may be only 1/10th, or even 1/100th of the former—the exact number may be similar to 18 000 that Leitman guessed.
Some people also hold the opposite view. They believe that the brain does not code concepts through a small group of neurons, but it is coded in a distributed manner, that is, many neurons participate together because if each concept uses tens of thousands of When neurons code, the brain may not have enough neurons to express all the concepts and how these concepts change. For example, if we can have more neurons in our brains, even in sparsely coded ways, we can encode grandma's smiles, darn clothes, tea or the appearance of people at bus stops, and the Queen's greetings to the public. And Skywalker Luke had a fight with Darth Vader on Tatooine in his childhood.
1301
1302
In the brain, how is memory coded? Neuroscientists have proposed two opposing theories, but they have not been conclusive. One theory is that every memory—such as the image of Skywalker Luke—is scattered and stored in millions or even billions of neurons. In recent years, another theory has been recognized by more scientists. This theory holds that the neuron's encoding of memory is "sparse" and that about a few thousand neurons can represent an image. When Luke's image appears, each of these neurons will be excited regardless of distance. A part (but not all) of this group of neurons will also be excited about the image of Yoda, another character related to Luke. Similarly, another group of neurons will be excited about the image of the actress Jennifer Aniston.
In order to answer this question, we must first consider that the concept that one can remember usually does not exceed 10,000. Compared with about 1 billion neurons in the medial temporal lobe, the 10,000 concept is not much. In addition, we have reason to believe that sparse coding and storage of concepts is very efficient. The neurons in the medial temporal lobe do not care about the different situations of a concept. For example, they do not care whether Luke is standing or sitting. They only care if the input information is related to Luke. These neurons only discharge the concept itself, and have nothing to do with the concrete manifestation of the concept. The abstraction of concepts - neurons can discharge all the manifestations of the concept of "Luke", reducing the amount of information that neurons need to encode, and making neurons highly selective, such as only discharging Luke, and Will not discharge to Jennifer.
Weidu's simulation study further developed this view. Based on a detailed model of visual information processing, Weedu simulated a neural network through a computer program that can recognize a variety of unmarked images such as airplanes, cars, motorcycles, and human faces. This set of procedures does not require the teacher's guidance to recognize the concepts expressed in the pictures, and no one tells it "This is an airplane. It is a truck." It must use the premises assumptions to independently identify. The hypothesis given to it is that although there are many images, they are actually different manifestations of a few people or things. Each person or thing is represented by a small group of neurons, just like we are in the medial temporal lobe. Discovered that way. After adding this sparse coding method in the software simulation, the neural network learns to distinguish different pictures of the same person or object. Even if these pictures have very large differences, the neural network can correctly distinguish. The results of this simulation study are very similar to those we obtained by recording the discharge of neurons in the human brain.
The association between concept cells
How does the brain represent the information of the outside world and how does it turn feeling into memory? This issue is closely related to our research. Take a look at a famous case (named H. M.) who suffers from intractable epilepsy. In order to control his strong epileptic symptoms, the neurosurgeon desperation had to choose to remove his hippocampus and both sides of the brain. The area connected to the hippocampus. After the operation, the patient can still identify other people and objects and can recall some things that were known before the operation. However, unexpectedly, he can no longer form new permanent memories. With the loss of the hippocampus, he would soon forget what he had just experienced, like the protagonist with a similar neurological disorder in Memento.
The above patient's story shows that the hippocampus (even the entire medial temporal lobe) is not necessary for perception, but for short-term memory (short duration) to long-term memory (lasting hours, days, or even years) The change is essential. We believe that “conceptual cells†located in the medial temporal region play a key role in the process of remembering the changes in what we are aware of (ie, externally input sensory information or brain recalls), long-term memory. It will then be stored in other areas of the cerebral cortex. We think that for the patient, when he was recognizing or retrieving Aniston, "Jennifer Aniston Neuron" was not necessary, but the patient wanted to put "Aniston" on her own. In the mind, establishing a connection or memory related to this female actress, this neuron is crucial - for example, he will remember the photo he had seen Aniston in the future.
Our brain may represent multiple forms of a thing as a unique concept through a few "conceptual cells." This representation requires only a small group of neurons and does not change as the specific form of the thing changes. The role of "conceptual cells" is very helpful in explaining our memory process. We will recall the overall image of Jennifer or Luke, not every detail of their faces. We do not need (and may not) recall the full details of everyone or everything that we encountered.
It is important to grasp key information about people and things that are relevant to us in a particular scene, rather than remembering a lot of meaningless details. If we happen to meet an acquaintance at the coffee shop, it is more important for us to remember some of the important things that happened after this meeting, not the person's clothes, or every word he said, not even coffee. The look of other strangers. "Concept cells" tend to be excited about things that are related to individuals because we usually remember things that are related to people or things that we are familiar with, without wasting energy to remember things that have nothing to do with us.
Memory is not just an isolated concept. The memory of Jennifer Aniston contains a series of stories about her and her role in films such as Friends. A complete recollection of a memory episode requires a connection between different but related concepts, such as the concept of “Jennifer Aniston†and “sitting on the sofa while watching “Friends†while eating Associated with concepts such as ice cream.
If two concepts are related, some of the neurons that encode one of the concepts may also be excited about the other. This explains the physiological process in which the brain's neurons code the interconnected things. The neurons will discharge other related concepts, which may be the formation of episodic memories (such as a series of events that occur after a coffee shop meets acquaintances) and flow of consciousness (awareness of content spontaneously from a concept Jump to the base of another concept. When we saw Jennifer Aniston, visual perception inspired our memories of concepts such as television, sofas, and ice cream. These interrelated concepts formed the memory of “watching the “Friends†episodeâ€. Different aspects of the same concept (stored in different brain regions) may also be related in a similar way, linking the bouquet, shape, color, and texture of a rose, or Jennifer's appearance and voice .
Since the superiority of storing advanced memories in the form of abstract concepts has obvious superiority, we must further explore why the representation of these concepts only requires a small group of neurons in the medial temporal lobe. A number of simulation studies have shown that sparse coding methods are necessary to quickly form the connection between different concepts - this may be the answer.
The technical details of the simulation study are quite complex, but the principle is very simple. Take an example where we met an acquaintance at a coffee shop. If we use distributed coding instead of the opposite sparse coding to represent this person, we need every detail of this person. Use many neurons for coding. The distributed coding of the coffee shop itself requires a large number of additional neurons. If you want to connect this person to this coffee shop, you need to establish a connection between a large number of neurons that represent the various details of the two concepts. This has not considered the question of linking these two concepts with other more concepts. For example, this coffee shop looks like a comfortable bookstore, and the person it meets looks like another person we know.
Establishing such a connection in a distributed network is very slow and may lead to memory confusion. In contrast, establishing such a connection in a sparse network is quick and easy, requiring only a few neurons to discharge both concepts, thus establishing a small number of connections between groups of neurons representing each concept. Another advantage of sparse networks is that adding new concepts does not have a significant impact on other existing concepts in the network; it is difficult to separate one concept from another in a distributed network. To add a new concept , and even need to change the boundaries of the entire network.
"Concept cells" connect perception and memory, and represent semantic knowledge such as people, places, objects, and all meaningful concepts that make up our personal world through abstraction and sparse coding. They are the bricks that build the memory building and enable us to form memories of facts and events in our lives. Their clever coding method allows our minds to open up innumerable trivial details and extract meaningful things to form new memories and establish new connections between concepts. The "conceptual cell" encodes the most important part of our experience.
"Concept cells" are not very similar to "grandmother cells" envisioned by Leitman, but they are likely to be important material foundations for human cognitive abilities, as well as the hardware components of thinking and memory.
Mini Box,Intel Celeron Processor Mini Pc,Customizable Pc,Mini Pc Desktop
C&Q Technology (Guangzhou) Co.,Ltd. , https://www.gzcqteq.com