An overview of modern projects of large-scale modeling of brain activity
In recent years, the field of large-scale modeling of brain activity has begun to actively develop and an increasing number of mathematicians and neuroscientists are involved in it. In this review I will conduct a brief overview of the most famous and successful projects in this area. Also, in conclusion, I will describe my thoughts on the prospects and usefulness of the further development of projects of this kind.
One of the first projects in this field that was widely publicized and financed was the Blue Brain Project [1], which IBM started in the summer of 2005 in cooperation with the Swiss Federal Institute of Technology in Lausanne.
')
The purpose of the Blue Brain Project is the detailed modeling of individual neurons and the type of neocortex of the brain they form - neocortical columns. In the cortex, neurons are organized into elementary units — neocortical columns, having about 0.5 mm in diameter and 2 mm in height. Each such column contains about 10 thousand neurons with a complex, but ordered structure of communication between themselves and with external neurogroups relative to the column. The data on the morphology and dynamics of the activity of rat neurons and other data on the physiology of the neuron obtained over the past decades of research on nerve cells served as the factual basis for the simulation.
Within the framework of this project, the neuron model takes into account the differences between the types of neurons, the spatial geometry of the neurons, the distribution of ion channels over the cell membrane surface and other parameters of the prototype neurons. Developers of the model point out that the diversity of types of neurons combined into a neurogroup is very important for the realization of the cognitive functions of this group, with each type of neuron present in certain layers of the column, and the spatial arrangement, density and volume of distribution of neurons of different types serve as the basis for the orderly distribution of activity across the network generally. The model also takes into account that the exact form and structure of the neuron affect its electrical properties and the ability to connect with other neurons, and the electrical properties of the neuron are determined by the diversity of ion channels.
For the three-dimensional modeling of neurogroups in the Blue Brain Project, the IBM Blue Gene / L computer is used (Fig. 1), which allows you to simulate the propagation of electrical activity inside a neocortical clone in real time.
Fig. 1. Schematic architecture of the Blue Gene / L supercomputer
At the end of 2006, it was possible to model one column of the neocortex of a young rat, consisting of 10,000 biologically plausible models of neurons with approximately 3x107 synapses between them.
At the end of 2007, the completion of Phase I of the Blue Brain project was announced. The results of this phase are:
• a new model of the grid structure, which automatically, upon request, generates a neural network based on the biological data provided;
• a new process of simulation and self-regulation, which automatically and systematically checks and calibrates the model before each release, in order to more accurately match the biological nature;
• The first model of the neocortex column at the cellular level, built exclusively on biological data.
According to the authors of the project, the obtained cellular models of neurons and the model of the column as a whole allow us to directly relate the simulated processes of propagation of activity with similar processes in the biological column of the prototype.
The continuation of the Blue Brain project is the new IBM project “Cognitive Computing via Synaptronics and Supercomputing” (C2S2), the start of which was announced on November 20, 2008 [2]. The company announced the launch of a project to develop a fundamentally new computer architecture that reproduces the organization of interneuron connections (synapses) and neural networks of the mammalian brain. The United States Agency for Advanced Defense Research Programs (DARPA) participates in the financing of the project as part of the System of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) project. This circumstance explains the actual lack of details on the progress of this project in scientific periodicals.
At the center of all research on the C2S2 project is the synapse, which, thanks to its plasticity, ensures the formation of individual experience. It is planned to develop models of neural networks with the number and density of distribution of synapses, comparable to the corresponding parameters in living organisms. It is noted that the brain, rather, is not a neural network, but a synaptic network, and thinking is the result of a biochemical organization of the brain.
If the project succeeds, according to its participants, a new class of artificial cognitive systems will be born, a new paradigm of computational architecture with numerous practical applications in all areas of human activity.
One of the most striking projects on large-scale brain modeling was carried out at The Neuroscience Institute by Yevgeny Izhikevich and Gerald Edelman.
In 2007, they modeled the mammalian thalamocortical system based on data on the human brain [3]. This model simulates the work of a million spike neurons that are calibrated to repeat the behavior of known types of neurons observed in vitro in the rat brain.
Fig. 2. Simplified diagram of the chip structure of the laminar cortex (above) and the nuclei of the thalamus (below)
The neuron model is based on the feminological model proposed by Izhikevich [4]. In the process of modeling, 22 types of neural cells were used (Fig. 2), which are obtained by changing the parameters of the Izhikevich model. Almost half a billion synapses with corresponding receptors, short-term synaptic plasticity and long-term STDP plasticity were used to connect neurons. In fig. 3 shows the dynamic visualization of simulation results.
Fig. 3. Spreading waves in the Izhikevich model.
(Red dots indicate spikes of excitatory neurons, black ones - inhibitory ones)
At the end of this review, as I mentioned earlier, I would like to say a few words about the appropriateness of such a large-scale simulation.
In the projects presented above, the brain is considered as some autonomous structure that can exist separately from the rest of the organism and, moreover, from the environment. Thus, it becomes unclear to assess the quality of the results of modeling - in which case we will understand that it is successful? Before the simulated brain does not set any tasks, not placed in any environment. In fact, the need for goal-directed behavior and the achievement of an adaptive result are not considered in such projects. They are aimed only at a detailed reproduction of the physical structure observed in the brain of real animals. Most likely, the consideration of the task of purposeful behavior in such projects is impossible, since with exact repetition of the physical structure of the brain, we still cannot determine what kind of experience it has and what tasks it can solve.
Participants of the Blue Brain project, in particular, argue that the development of their research will help in creating AI in a fairly short period of time (the next 20-25 years). This statement sounds at least
loud enough, but there is one fact that does not allow to believe in it. Basically, these studies are aimed at studying the spread of activity in the brain and modeling of rhythms. However, within these projects, almost no attention is paid to training, which most likely negates the usefulness of developments in this area as a basis for creating AI.
[1] . Markram H. "The blue brain project". // Nat Rev Neurosci. Vol. 7, pp. 153-160 (2006).
[2] . IBM Pressroom [Electronic resource] / "IBM Seeks to Build the Computer of the Future Based on Insights from the Brain" - www-03.ibm.com/press/us/en/pressrelease/26123.wss#release
[3] . Izhikevich E., Edelman G. "Large-scale model of mammalian thalamocortical systems". // PNAS. Vol. 105, no.9, pp. 3593-3598 (2008).
[4] . Izhikevich E. "The Simple Model of Spiking Neurons". // IEEE Transactions on Neural Networks. Vol. 14, no. 6 (2003).
Large-scale models of brain activity
One of the first projects in this field that was widely publicized and financed was the Blue Brain Project [1], which IBM started in the summer of 2005 in cooperation with the Swiss Federal Institute of Technology in Lausanne.
')
The purpose of the Blue Brain Project is the detailed modeling of individual neurons and the type of neocortex of the brain they form - neocortical columns. In the cortex, neurons are organized into elementary units — neocortical columns, having about 0.5 mm in diameter and 2 mm in height. Each such column contains about 10 thousand neurons with a complex, but ordered structure of communication between themselves and with external neurogroups relative to the column. The data on the morphology and dynamics of the activity of rat neurons and other data on the physiology of the neuron obtained over the past decades of research on nerve cells served as the factual basis for the simulation.
Within the framework of this project, the neuron model takes into account the differences between the types of neurons, the spatial geometry of the neurons, the distribution of ion channels over the cell membrane surface and other parameters of the prototype neurons. Developers of the model point out that the diversity of types of neurons combined into a neurogroup is very important for the realization of the cognitive functions of this group, with each type of neuron present in certain layers of the column, and the spatial arrangement, density and volume of distribution of neurons of different types serve as the basis for the orderly distribution of activity across the network generally. The model also takes into account that the exact form and structure of the neuron affect its electrical properties and the ability to connect with other neurons, and the electrical properties of the neuron are determined by the diversity of ion channels.
For the three-dimensional modeling of neurogroups in the Blue Brain Project, the IBM Blue Gene / L computer is used (Fig. 1), which allows you to simulate the propagation of electrical activity inside a neocortical clone in real time.
Fig. 1. Schematic architecture of the Blue Gene / L supercomputer
At the end of 2006, it was possible to model one column of the neocortex of a young rat, consisting of 10,000 biologically plausible models of neurons with approximately 3x107 synapses between them.
At the end of 2007, the completion of Phase I of the Blue Brain project was announced. The results of this phase are:
• a new model of the grid structure, which automatically, upon request, generates a neural network based on the biological data provided;
• a new process of simulation and self-regulation, which automatically and systematically checks and calibrates the model before each release, in order to more accurately match the biological nature;
• The first model of the neocortex column at the cellular level, built exclusively on biological data.
According to the authors of the project, the obtained cellular models of neurons and the model of the column as a whole allow us to directly relate the simulated processes of propagation of activity with similar processes in the biological column of the prototype.
The continuation of the Blue Brain project is the new IBM project “Cognitive Computing via Synaptronics and Supercomputing” (C2S2), the start of which was announced on November 20, 2008 [2]. The company announced the launch of a project to develop a fundamentally new computer architecture that reproduces the organization of interneuron connections (synapses) and neural networks of the mammalian brain. The United States Agency for Advanced Defense Research Programs (DARPA) participates in the financing of the project as part of the System of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) project. This circumstance explains the actual lack of details on the progress of this project in scientific periodicals.
At the center of all research on the C2S2 project is the synapse, which, thanks to its plasticity, ensures the formation of individual experience. It is planned to develop models of neural networks with the number and density of distribution of synapses, comparable to the corresponding parameters in living organisms. It is noted that the brain, rather, is not a neural network, but a synaptic network, and thinking is the result of a biochemical organization of the brain.
If the project succeeds, according to its participants, a new class of artificial cognitive systems will be born, a new paradigm of computational architecture with numerous practical applications in all areas of human activity.
One of the most striking projects on large-scale brain modeling was carried out at The Neuroscience Institute by Yevgeny Izhikevich and Gerald Edelman.
In 2007, they modeled the mammalian thalamocortical system based on data on the human brain [3]. This model simulates the work of a million spike neurons that are calibrated to repeat the behavior of known types of neurons observed in vitro in the rat brain.
Fig. 2. Simplified diagram of the chip structure of the laminar cortex (above) and the nuclei of the thalamus (below)
The neuron model is based on the feminological model proposed by Izhikevich [4]. In the process of modeling, 22 types of neural cells were used (Fig. 2), which are obtained by changing the parameters of the Izhikevich model. Almost half a billion synapses with corresponding receptors, short-term synaptic plasticity and long-term STDP plasticity were used to connect neurons. In fig. 3 shows the dynamic visualization of simulation results.
Fig. 3. Spreading waves in the Izhikevich model.
(Red dots indicate spikes of excitatory neurons, black ones - inhibitory ones)
Conclusion
At the end of this review, as I mentioned earlier, I would like to say a few words about the appropriateness of such a large-scale simulation.
In the projects presented above, the brain is considered as some autonomous structure that can exist separately from the rest of the organism and, moreover, from the environment. Thus, it becomes unclear to assess the quality of the results of modeling - in which case we will understand that it is successful? Before the simulated brain does not set any tasks, not placed in any environment. In fact, the need for goal-directed behavior and the achievement of an adaptive result are not considered in such projects. They are aimed only at a detailed reproduction of the physical structure observed in the brain of real animals. Most likely, the consideration of the task of purposeful behavior in such projects is impossible, since with exact repetition of the physical structure of the brain, we still cannot determine what kind of experience it has and what tasks it can solve.
Participants of the Blue Brain project, in particular, argue that the development of their research will help in creating AI in a fairly short period of time (the next 20-25 years). This statement sounds at least
loud enough, but there is one fact that does not allow to believe in it. Basically, these studies are aimed at studying the spread of activity in the brain and modeling of rhythms. However, within these projects, almost no attention is paid to training, which most likely negates the usefulness of developments in this area as a basis for creating AI.
Literature
[1] . Markram H. "The blue brain project". // Nat Rev Neurosci. Vol. 7, pp. 153-160 (2006).
[2] . IBM Pressroom [Electronic resource] / "IBM Seeks to Build the Computer of the Future Based on Insights from the Brain" - www-03.ibm.com/press/us/en/pressrelease/26123.wss#release
[3] . Izhikevich E., Edelman G. "Large-scale model of mammalian thalamocortical systems". // PNAS. Vol. 105, no.9, pp. 3593-3598 (2008).
[4] . Izhikevich E. "The Simple Model of Spiking Neurons". // IEEE Transactions on Neural Networks. Vol. 14, no. 6 (2003).
Source: https://habr.com/ru/post/101787/
All Articles