-
Categories
-
Pharmaceutical Intermediates
-
Active Pharmaceutical Ingredients
-
Food Additives
- Industrial Coatings
- Agrochemicals
- Dyes and Pigments
- Surfactant
- Flavors and Fragrances
- Chemical Reagents
- Catalyst and Auxiliary
- Natural Products
- Inorganic Chemistry
-
Organic Chemistry
-
Biochemical Engineering
- Analytical Chemistry
-
Cosmetic Ingredient
- Water Treatment Chemical
-
Pharmaceutical Intermediates
Promotion
ECHEMI Mall
Wholesale
Weekly Price
Exhibition
News
-
Trade Service
Xinzhiyuan report source: cell editor: Emil, LQ [Xinzhiyuan Guide] The latest issue of Cell published a cross-border study conducted by Chinese scientists Chen Guang and Dr.
Li Nuo from Baylor College of Medicine in cooperation with Stanford University: passed Explore the biological mechanism of brain nerve activity to improve the robustness of artificial neural networks
.
In a sense, the machine learning method widely used in the AI field was inspired by neurophysiology and invented
.
Although the calculation methods and the latest achievements of machine learning are completely different from the neural structure of the biological brain, people still like to constantly compare machine learning with the biological brain
.
For example, the three fathers of deep learning, Hinton, LeCun, and Bengio, jointly published an article the day before yesterday to discuss the potential of deep learning in the future.
They are also comparing deep learning with humans.
Among them, an important advantage of the human brain is that it still faces new environments.
It has excellent robustness (robustness) and will not be disturbed by complex environments and noise
.
(Click on the picture to enter the article link) For machine learning, constantly changing environmental parameters and noise will cause huge interference to the calculation, resulting in a decrease in the accuracy of the results
.
How does the human or other animal brain maintain robustness in the face of new environmental changes and noise? The "short-term memory" function possessed by humans or animals is a manifestation of the robustness of the brain
.
For example, at a dance party, you can temporarily remember the name and mobile phone number of the stranger who greeted you; at the supermarket, you can also remember the shopping list you saw before
.
A few days ago, the latest article published on Cell, starting from exploring the neural mechanism of the brain, uncovered the mystery.
.
On July 1, Dr.
Guang Chen and Dr.
Li Nuo from Baylor College of Medicine in the United States and PhD students Byungwoo Kang and Dr.
Shaul Druckmann from Stanford University published an article titled "Modularity and Modularity of Frontal Cortex Networks" in Cell.
"Robustness" (Modularity and robustness of frontal cortical networks) article
.
What does the paper say? This paper mainly describes the following characteristics: The cross-frontal cortex network is related to the selection and the variability of the redundant representation of the individual shows that the modular organization is necessary for robustness.
Dynamic gating allows the brain area Communication and separation of redundant modular representations naturally appear in robust network models.
The neural activity on which short-term memory is based is maintained by a network of interconnected brain regions
.
At present, we still don't know how brain regions interact to maintain continuous activity, while showing robustness to the damage of some information in the network
.
The researchers also measured the activity of large groups of neurons in the frontal lobe of mice to detect interactions between brain regions
.
Activities across the cerebral hemispheres are coordinated to maintain coherent short-term memory
.
Throughout the mouse experiment, the researchers found individual differences in the network organization of the frontal cortex
.
The robustness of continuous activity to disturbance requires a modular organization: each hemisphere retains continuous activity when the other hemisphere is disturbed, thereby preventing the spread of local disturbances
.
A dynamic gating mechanism allows the brain hemispheres to coordinate information while simultaneously gating out damaged information
.
The results of this study indicate that strong short-term memory is mediated by the representation of redundant modules across brain regions
.
The redundant module representation will naturally appear in the neural network model that learns robust and dynamic
.
Test results The mice performed a delayed response task.
They used their whiskers to identify the location (front or back) of the tactile stimulus, and reported the position of the object through directional licking ("licking left" or "licking right") to obtain water Rewards
.
There is a delay time (1.
7 seconds) between the sensory command and the behavioral response
.
This requires mice to use short-term memory to produce the correct selection response (Figure 1A; STAR Methods)
.
In this study, when the mice were performing short-term memory behavioral tasks, they also recorded large-scale "neuronal population electrical activities" and "optogenetic disturbances" in the prefrontal cortex on both sides of their brains
.
The experiment found that the prefrontal cortex on both sides simultaneously represented two similar short-term memory information, and the neural activity that represented this information showed a high degree of "consistency"
.
Further experiments and video analysis of mouse movement suggest that the coordination of activities on both sides is likely to be formed by the interaction of the neural networks on both sides, rather than the common input of other brain regions
.
To verify this, the researchers directly measured the interaction between the brain regions on both sides
.
In the early stage of memory, optogenetic methods were used to directly inhibit the neural activity of the unilateral prefrontal cortex, while observing the degree of influence of the disturbance on the contralateral brain area
.
It was found that some mice showed a high degree of modularity, that is, inhibiting the nerve activity on the left side had almost no effect on the right side, and inhibiting the nerve activity on the right side had almost no effect on the left side, showing two modules
.
Other mice lack modularity.
The left side strongly dominates the activities on the right side while the right side does not affect the left side, presenting an asymmetric single module
.
In general, different mice have different degrees of modularity, showing approximately continuous changes
.
Further analysis revealed that the strongly asymmetrical single module showed a high degree of consistency in the activities on both sides, while the relatively independent dual-module system showed a lower consistency
.
It shows that the coordination of activities on both sides is indeed related to the interaction of the cortex on both sides
.
Further experiments found that this differentiated and left-sided dominance interaction between the two sides of the brain in different mice is not completely innate.
Changing the structure of behavioral tasks can reverse the left-side dominance to the right-side dominance
.
The above results indicate that the low-interaction modularity of the brain regions is necessary for robustness to prevent damage to the spread of information (modularity), but at the same time, the normal memory information that does not disturb the brain regions needs to be remedied by the strong interaction of the brain regions ( Error-correction (error-correction) disturbs the information that has been destroyed in the brain area
.
Finally, researchers train recurrent neural networks (RNN) to perform the same behavioral tasks to explore what kind of network architecture can be the solution to form a robust network, and under what conditions can be repeated in artificially trained neural networks Develop the phenomena and principles observed in the experiment
.
Studies have shown that for a neural network to have a robust performance in this short-term memory task, three basic conditions need to be met, namely, the modularized initial connection between the cortical networks on both sides, modularized training, and the nonlinear characteristics of neurons
.
In this trained neural network, the state-dependent gating phenomenon between the networks on both sides also emerges at the same time
.
And similar to experimental research, changing the relative intensity of the network sensory input on both sides of the training can change the symmetry structure of the artificial neural network
.
Machine learning may achieve new breakthroughs because of this research
.
Reference: https:// :// past-
Li Nuo from Baylor College of Medicine in cooperation with Stanford University: passed Explore the biological mechanism of brain nerve activity to improve the robustness of artificial neural networks
.
In a sense, the machine learning method widely used in the AI field was inspired by neurophysiology and invented
.
Although the calculation methods and the latest achievements of machine learning are completely different from the neural structure of the biological brain, people still like to constantly compare machine learning with the biological brain
.
For example, the three fathers of deep learning, Hinton, LeCun, and Bengio, jointly published an article the day before yesterday to discuss the potential of deep learning in the future.
They are also comparing deep learning with humans.
Among them, an important advantage of the human brain is that it still faces new environments.
It has excellent robustness (robustness) and will not be disturbed by complex environments and noise
.
(Click on the picture to enter the article link) For machine learning, constantly changing environmental parameters and noise will cause huge interference to the calculation, resulting in a decrease in the accuracy of the results
.
How does the human or other animal brain maintain robustness in the face of new environmental changes and noise? The "short-term memory" function possessed by humans or animals is a manifestation of the robustness of the brain
.
For example, at a dance party, you can temporarily remember the name and mobile phone number of the stranger who greeted you; at the supermarket, you can also remember the shopping list you saw before
.
A few days ago, the latest article published on Cell, starting from exploring the neural mechanism of the brain, uncovered the mystery.
.
On July 1, Dr.
Guang Chen and Dr.
Li Nuo from Baylor College of Medicine in the United States and PhD students Byungwoo Kang and Dr.
Shaul Druckmann from Stanford University published an article titled "Modularity and Modularity of Frontal Cortex Networks" in Cell.
"Robustness" (Modularity and robustness of frontal cortical networks) article
.
What does the paper say? This paper mainly describes the following characteristics: The cross-frontal cortex network is related to the selection and the variability of the redundant representation of the individual shows that the modular organization is necessary for robustness.
Dynamic gating allows the brain area Communication and separation of redundant modular representations naturally appear in robust network models.
The neural activity on which short-term memory is based is maintained by a network of interconnected brain regions
.
At present, we still don't know how brain regions interact to maintain continuous activity, while showing robustness to the damage of some information in the network
.
The researchers also measured the activity of large groups of neurons in the frontal lobe of mice to detect interactions between brain regions
.
Activities across the cerebral hemispheres are coordinated to maintain coherent short-term memory
.
Throughout the mouse experiment, the researchers found individual differences in the network organization of the frontal cortex
.
The robustness of continuous activity to disturbance requires a modular organization: each hemisphere retains continuous activity when the other hemisphere is disturbed, thereby preventing the spread of local disturbances
.
A dynamic gating mechanism allows the brain hemispheres to coordinate information while simultaneously gating out damaged information
.
The results of this study indicate that strong short-term memory is mediated by the representation of redundant modules across brain regions
.
The redundant module representation will naturally appear in the neural network model that learns robust and dynamic
.
Test results The mice performed a delayed response task.
They used their whiskers to identify the location (front or back) of the tactile stimulus, and reported the position of the object through directional licking ("licking left" or "licking right") to obtain water Rewards
.
There is a delay time (1.
7 seconds) between the sensory command and the behavioral response
.
This requires mice to use short-term memory to produce the correct selection response (Figure 1A; STAR Methods)
.
In this study, when the mice were performing short-term memory behavioral tasks, they also recorded large-scale "neuronal population electrical activities" and "optogenetic disturbances" in the prefrontal cortex on both sides of their brains
.
The experiment found that the prefrontal cortex on both sides simultaneously represented two similar short-term memory information, and the neural activity that represented this information showed a high degree of "consistency"
.
Further experiments and video analysis of mouse movement suggest that the coordination of activities on both sides is likely to be formed by the interaction of the neural networks on both sides, rather than the common input of other brain regions
.
To verify this, the researchers directly measured the interaction between the brain regions on both sides
.
In the early stage of memory, optogenetic methods were used to directly inhibit the neural activity of the unilateral prefrontal cortex, while observing the degree of influence of the disturbance on the contralateral brain area
.
It was found that some mice showed a high degree of modularity, that is, inhibiting the nerve activity on the left side had almost no effect on the right side, and inhibiting the nerve activity on the right side had almost no effect on the left side, showing two modules
.
Other mice lack modularity.
The left side strongly dominates the activities on the right side while the right side does not affect the left side, presenting an asymmetric single module
.
In general, different mice have different degrees of modularity, showing approximately continuous changes
.
Further analysis revealed that the strongly asymmetrical single module showed a high degree of consistency in the activities on both sides, while the relatively independent dual-module system showed a lower consistency
.
It shows that the coordination of activities on both sides is indeed related to the interaction of the cortex on both sides
.
Further experiments found that this differentiated and left-sided dominance interaction between the two sides of the brain in different mice is not completely innate.
Changing the structure of behavioral tasks can reverse the left-side dominance to the right-side dominance
.
The above results indicate that the low-interaction modularity of the brain regions is necessary for robustness to prevent damage to the spread of information (modularity), but at the same time, the normal memory information that does not disturb the brain regions needs to be remedied by the strong interaction of the brain regions ( Error-correction (error-correction) disturbs the information that has been destroyed in the brain area
.
Finally, researchers train recurrent neural networks (RNN) to perform the same behavioral tasks to explore what kind of network architecture can be the solution to form a robust network, and under what conditions can be repeated in artificially trained neural networks Develop the phenomena and principles observed in the experiment
.
Studies have shown that for a neural network to have a robust performance in this short-term memory task, three basic conditions need to be met, namely, the modularized initial connection between the cortical networks on both sides, modularized training, and the nonlinear characteristics of neurons
.
In this trained neural network, the state-dependent gating phenomenon between the networks on both sides also emerges at the same time
.
And similar to experimental research, changing the relative intensity of the network sensory input on both sides of the training can change the symmetry structure of the artificial neural network
.
Machine learning may achieve new breakthroughs because of this research
.
Reference: https:// :// past-
