A spiking neural network simulation system and method integrating brain glial cells

Abstract

The present invention relates to a spiking neural network simulation system and method for fused glial cells. The spiking neural network simulation system for fused glial cells is provided with a simulation software framework and a neuromorphic computer. By using the simulation software framework to interact with the neuromorphic computer for data, the simulation of the spiking neural network fused with glial cells is achieved. This can expand the application scope and intelligent computing capabilities of the neuromorphic computer, reduce the difficulty and development cost of constructing brain-like neural networks, and thus better serve fields such as brain science research, artificial intelligence, and brain-like computing.

Description

Technical Field

[0001] The present invention relates to the field of intelligent computing technology, and in particular to a pulse neural network simulation system and method integrating brain glial cells. Background Art

[0002] Neuromorphic computing, also known as brain-inspired computing and neuromorphic computing, simulates the neuronal and synaptic connectivity and pulse-like signal transmission of the human brain. It redistributes neurons through a novel network model, Spiking Neural Networks (SNNs), in an effort to achieve brain-like natural intelligence. Characterized by integrated storage and computing, event-driven operation, and high parallelism, neuromorphic computing has been called a key direction for the next generation of artificial intelligence. Neuromorphic computers, AI computers that utilize neuromorphic computing, employ a novel computing architecture and boast low power consumption and high robustness. Neuromorphic computing has become a hot topic in the field of artificial intelligence, with increasing attention being paid to the role and potential of glial cells in brain information integration and processing. A growing body of research indicates that glial cells not only play a supporting and safeguarding role in brain activity but are also deeply involved in information integration and processing, making them a key component of brain intelligence. However, most current neuromorphic computers and their spiking neural network development software only provide two neural network structural units: neurons and synapses. That is, they can only build and simulate spiking neural networks composed of neurons and synapses, but cannot systematically and completely build and simulate Blended Glial Cell's Spiking Neural Networks (BGSNN). This limits the application scope and intelligent computing capabilities of neuromorphic computers. Summary of the Invention

[0003] The purpose of the present invention is to provide a pulse neural network simulation system and method that integrates brain glial cells, which can improve the application scope and intelligent computing capabilities of neuromorphic computers and reduce the difficulty and development cost of constructing brain-like neural networks.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A spiking neural network simulation system integrating brain glial cells, comprising: a simulation software framework and a neuromorphic computer;

[0006] The simulation software framework interacts with the neuromorphic computer for data exchange; the simulation software framework is used to obtain the pulse neural network of the fused brain glial cells, and to generate a neural network driver and neural network driver configuration data based on the pulse neural network of the fused brain glial cells; the neuromorphic computer is used to realize the simulation of the pulse neural network of the fused brain glial cells based on the neural network driver and the neural network driver configuration data.

[0007] Preferably, the simulation software framework includes: a neuron dynamics component, a brain glial cell dynamics component, a neural cell connection matrix component, a BGSNN component manager, a driver library component, a neuromorphic computer driver manager, and a custom model creation tool;

[0008] The neuron dynamics component, the brain glial cell dynamics component and the neural cell connection matrix component all interact with the BGSNN component manager for data; the neuromorphic computer driver manager is connected to the neuron dynamics component, the brain glial cell dynamics component and the neural cell connection matrix component respectively; the neuromorphic computer driver manager interacts with the driver library component for data; the driver library component interacts with the neuromorphic computer for data; the driver library component is connected to the custom model creation tool;

[0009] The neuron dynamics component is embedded with neuron pulse receptor dynamics, neuron pulse dendrite conduction dynamics, neuron membrane potential update program and neuron axon pulse emission program, so as to process and respond to the transmitted neuron pulse signals in the time domain;

[0010] The brain glial cell dynamics component is implanted with neuronal ion information receptor dynamics, glial gap information receptor dynamics, neuronal ion conduction dynamics, glial gap conduction dynamics, brain glial cell membrane potential update program, brain glial cell glial protrusion information release program, brain glial cell glial gap information release program and brain glial cell glial ion information release program, so as to process and respond to the conducted neuronal ion information and glial gap information in the time domain;

[0011] The neural cell connection matrix component is used to construct connections between neural cell populations in a spiking neural network;

[0012] The BGSNN component manager is used to obtain and parse the pulse neural network instance of the fused brain glial cells, and to select corresponding implementation modules from the neuron dynamics component, the brain glial cell dynamics component, and the neural cell connection matrix component to generate network configuration parameters and network initial state parameters corresponding to the pulse neural network instance of the fused brain glial cells;

[0013] The driver library component is used to simulate the nodes of the pulse neural network instance of the fused brain glial cells, to create the connection matrix data of the pulse neural network instance of the fused brain glial cells, and to realize data communication with the neuromorphic computer;

[0014] The custom model creation tool is used to provide a standardized method for creating a custom network node model;

[0015] The neuromorphic computer driver manager is used to call the correct hardware driver from the driver library component and select and download the correct neural network driver to the neuromorphic computer according to the pulse neural network instance of the fused brain glial cells and the neuromorphic computer model, and drive the neuromorphic computer to complete the simulation of the pulse neural network instance of the fused brain glial cells.

[0016] Preferably, the driver library component includes: a network node driver library component, a network connection driver library component and a hardware driver library component;

[0017] The network node driver library component is connected to the custom model creation tool;

[0018] The network node driver library component includes a neuron model driver and a glial cell model driver running on the neuromorphic computer; the network node driver library component is used to simulate a node of a pulse neural network instance of fused glial cells input by a user on the neuromorphic computer;

[0019] The network connection driver library component includes static connection creation and dynamic connection creation programs running on the neuromorphic computer; the network connection driver library component is used to create connection matrix data in the pulse neural network instance of the fused glial cells input by the user on the neuromorphic computer;

[0020] The hardware driver library component includes an IO interface and a hardware driver; the IO interface communicates data with the neuromorphic computer; the hardware driver is used to implement the downloading of network configuration parameters and applications and the uploading of running process data and recorded data.

[0021] Preferably, the connections between neural cell populations in the spiking neural network include synaptic connections, neuronal ionic connections, glial gap junctions, glial process connections and glial ionic connections.

[0022] Preferably, the standardized method for creating a customized network node model is based on a neural network node driver editor provided by the manufacturer, and provides a standardized editing template based on the open source PyNN interface, dividing the neural network nodes into four parts, namely: receptor dynamics, conduction dynamics, membrane potential dynamics and information release.

[0023] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0024] The pulse neural network simulation system for integrating brain glial cells provided by the present invention is provided with a simulation software framework and a neuromorphic computer. By adopting the simulation software framework to interact with the neuromorphic computer for data, the simulation of the pulse neural network integrating brain glial cells can be realized, which can improve the application scope and intelligent computing ability of the neuromorphic computer, reduce the difficulty and development cost of constructing brain-like neural networks, and thus better serve the fields of brain science research, artificial intelligence, brain-like computing, etc.

[0025] The present invention also provides a spiking neural network simulation method for fused glial cells, which is applied to the spiking neural network simulation system for fused glial cells provided above; the spiking neural network simulation method for fused glial cells comprises:

[0026] Obtain an example of a spiking neural network that fuses brain glial cells;

[0027] Determining whether there is a neuron network node in the pulse neural network instance of the fused glial cells to obtain a first determination result;

[0028] When the first determination result is that a neuron network node exists in the pulse neural network instance of the fused glial cells, after generating neuron model data, determining whether a glial network node exists in the pulse neural network instance of the fused glial cells to obtain a second determination result; the neuron model data includes: pulse receptor dynamics data, dendritic conduction dynamics data, neuron membrane potential dynamics data, and neuron axon pulse discharge data;

[0029] When the first determination result is that there is no neuron network node in the pulse neural network instance of the fused glial cells, directly determining whether there is a glial network node in the pulse neural network instance of the fused glial cells to obtain a second determination result;

[0030] When the second determination result is that a glial network node exists in the pulse neural network instance of the fused glial cells, after generating the glial cell model data, determining whether there is a connection between the neural cell populations in the pulse neural network instance of the fused glial cells, and obtaining a third determination result; the glial cell model data includes: neuron ion information receptor dynamics data, glial gap information receptor dynamics data, glial cell membrane potential dynamics data, glial cell glial process information release data, glial cell glial gap information release data, and glial cell glial ion information release data;

[0031] When the second determination result is that there is no glial network node in the spiking neural network instance of the fused glial cells, directly determining whether there is a connection between the nerve cell populations in the spiking neural network instance of the fused glial cells to obtain a third determination result;

[0032] When the third determination result indicates that there is a connection between the nerve cell populations in the spiking neural network instance of the fused glial cells, after generating network connection application data, calling a driver to complete the simulation of the spiking neural network instance of the fused glial cells; the network connection application data includes: glial gap connection data, neuron ion connection data, synaptic connection data, glial process connection data, and glial ion connection data;

[0033] When the third determination result is that there is no connection between the nerve cell populations in the pulse neural network instance of the fused glial cells, the driver is directly called to complete the simulation of the pulse neural network instance of the fused glial cells.

[0034] Preferably, the process of generating neuron model data includes:

[0035] Reading a neuron pulse receptor dynamics user configuration parameter, and calling a neuron pulse dynamics parameter generation method corresponding to the neuron pulse receptor dynamics user configuration parameter to generate pulse receptor dynamics data;

[0036] Reading a neuron pulse dendrite conduction dynamics user configuration parameter, and calling a neuron pulse dendrite conduction dynamics parameter generation method corresponding to the neuron pulse dendrite conduction dynamics user configuration parameter to generate dendrite conduction dynamics data;

[0037] Reading neuronal membrane potential dynamics user configuration parameters, calling a neuronal membrane potential dynamics parameter generation method corresponding to the neuronal membrane potential dynamics user configuration parameters to generate neuronal membrane potential dynamics data;

[0038] The neuron axon hillock pulse emission user configuration parameters are read, and a neuron axon hillock pulse emission parameter generation method corresponding to the neuron axon hillock pulse emission user configuration parameters is called to generate neuron axon hillock pulse emission data.

[0039] Preferably, the process of generating brain glial cell model data includes:

[0040] Read the neuron ion information receptor dynamics user configuration parameters, call the neuron ion information receptor dynamics parameter generation method corresponding to the neuron ion information receptor dynamics user configuration parameters to generate neuron ion information receptor dynamics data; read the glial gap information receptor dynamics user configuration parameters, call the glial gap information receptor dynamics parameter generation method corresponding to the glial gap information receptor dynamics user configuration parameters to generate glial gap information receptor dynamics data; read the brain glial cell membrane potential dynamics user configuration parameters, call the brain glial cell membrane potential dynamics parameter generation method corresponding to the brain glial cell membrane potential user configuration parameters to generate brain glial cell membrane potential dynamics data; read the glial processes of the brain glial cells The information is released by the user configuration parameters, and the method for generating the glial cell glial protrusion information release parameters corresponding to the glial cell glial protrusion information release user configuration parameters is called to generate the glial cell glial protrusion information release data; the glial cell glial gap information release user configuration parameters are read, and the method for generating the glial cell glial gap information release parameters corresponding to the glial cell glial gap information release user configuration parameters is called to generate the glial cell glial gap information release data; the glial cell glial ion information release user configuration parameters are read, and the method for generating the glial cell glial ion information release parameters corresponding to the glial cell glial ion information release user configuration parameters is called to generate the glial cell glial ion information release data.

[0041] Preferably, the process of generating network connection application data includes:

[0042] When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused glial cells is a synaptic network connection, reading the synaptic connection user configuration parameters, and calling the synaptic connection parameter generation method corresponding to the synaptic connection user configuration parameters to generate synaptic connection data;

[0043] When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused glial cells is a neuron ion connection, reading the neuron ion connection user configuration parameters, and calling the neuron ion connection parameter generation method corresponding to the neuron ion connection user configuration parameters to generate neuron ion connection data;

[0044] When the third determination result is that the connection between the nerve cell populations in the spiking neural network instance of the fused brain glial cells is a glial gap junction, reading the glial gap junction user configuration parameters, and calling the glial gap junction parameter generation method corresponding to the glial gap junction user configuration parameters to generate glial gap junction data;

[0045] When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused brain glial cells is a glial process connection, reading the glial process connection user configuration parameters, and calling the glial process connection parameter generation method corresponding to the glial process connection user configuration parameters to generate glial process connection data;

[0046] When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused brain glial cells is a glial ion connection, the glial ion connection user configuration parameters are read, and the glial ion connection parameter generation method corresponding to the glial ion connection user configuration parameters is called to generate glial ion connection data.

[0047] Preferably, the process of calling the driver to complete the simulation of the pulse neural network instance integrating brain glial cells includes:

[0048] After correspondingly copying the neuron model driver in the network node driver library component, the brain glial cell model driver in the network node driver library component, or the static connection creation and dynamic connection creation programs in the network connection driver library component based on the neuron model data, the brain glial cell model data, or the network connection application data, the IO interface in the hardware driver library component is called to complete data interaction with the neuromorphic computer to complete the simulation of the pulse neural network instance that integrates brain glial cells.

[0049] Since the technical effects achieved by the pulse neural network simulation method for fusion of brain glial cells provided by the present invention are the same as the technical effects achieved by the pulse neural network simulation system for fusion of brain glial cells provided above, they will not be described in detail here. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 This is a structural block diagram of the pulse neural network simulation system for integrating brain glial cells provided by the present invention;

[0052] Figure 2 This is a diagram of the implementation architecture of the pulse neural network simulation method for integrating brain glial cells provided by the present invention;

[0053] Figure 3 A flowchart for creating a custom multi-input and multi-output network node model driver provided by an embodiment of the present invention;

[0054] Figure 4 A schematic diagram of the composition of network node application data provided by an embodiment of the present invention;

[0055] Figure 5 A schematic diagram of the network connection data composition provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0057] The purpose of the present invention is to provide a pulse neural network simulation system and method that integrates brain glial cells, which can improve the application scope and intelligent computing capabilities of neuromorphic computers and reduce the difficulty and development cost of constructing brain-like neural networks.

[0058] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0059] The present invention provides a pulse neural network simulation system integrating brain glial cells, comprising: a simulation software framework and a neuromorphic computer.

[0060] The simulation software framework exchanges data with the neuromorphic computer. The simulation software framework is used to obtain the spiking neural network of the fused glial cells and to generate a neural network driver and neural network driver configuration data based on the spiking neural network of the fused glial cells. The neuromorphic computer is used to simulate the spiking neural network of the fused glial cells based on the neural network driver and neural network driver configuration data.

[0061] Among them, such as Figure 1 As shown, the simulation software framework provided by the present invention includes: a neuron dynamics component, a glial cell dynamics component, a neural cell connection matrix component, a BGSNN component manager, a driver library component, a neuromorphic computer driver manager, and a custom model creation tool. Among them, BGSNN is the abbreviation of spiking neural network integrated with glial cells.

[0062] The Neuron Dynamics component, the Glial Cell Dynamics component, and the Neural Cell Connectivity Matrix component all exchange data with the BGSNN component manager. The Neuromorphic Computer Driver Manager connects to the Neuron Dynamics component, the Glial Cell Dynamics component, and the Neural Cell Connectivity Matrix component, respectively. The Neuromorphic Computer Driver Manager also exchanges data with the Driver Library component. The Driver Library component also exchanges data with the Neuromorphic Computer. The Driver Library component connects to the custom model creation tool.

[0063] The neuron dynamics component is embedded with neuron pulse receptor dynamics, neuron pulse dendrite conduction dynamics, neuron membrane potential update program and neuron axon pulse emission program, so as to process and respond to the transmitted neuron pulse signals in the time domain.

[0064] The brain glial cell dynamics component is implanted with neuronal ion information receptor dynamics, glial gap information receptor dynamics, neuronal ion conduction dynamics, glial gap conduction dynamics, brain glial cell membrane potential update program, brain glial cell glial process information release program, brain glial cell glial gap information release program and brain glial cell glial ion information release program, which are used to process and respond to the transmitted neuronal ion information and glial gap information in the time domain.

[0065] The Neural Cell Connection Matrix component is used to construct connections between neural cell populations in a spiking neural network. These connections include synaptic connections, neuronal ionic connections, glial gap junctions, glial process connections, and glial ionic connections.

[0066] The BGSNN component manager is used to obtain and parse the spiking neural network instance of fused brain glial cells, and to select the corresponding implementation modules in the neuron dynamics component, brain glial cell dynamics component and neural cell connection matrix component to generate the network configuration parameters and network initial state parameters corresponding to the spiking neural network instance of fused brain glial cells.

[0067] The driver library component is used to simulate the nodes of the pulse neural network instance of the fused brain glial cells, to create the connection matrix data of the pulse neural network instance of the fused brain glial cells, and to realize data communication with the neuromorphic computer.

[0068] The driver library component includes: a network node driver library component, a network connection driver library component, and a hardware driver library component. The network node driver library component is connected to the custom model creation tool. The network node driver library component contains a neuron model driver and a glial cell model driver that run on the neuromorphic computer. The network node driver library component is used to simulate nodes of a spiking neural network instance of fused glial cells input by the user on the neuromorphic computer. The network connection driver library component contains static connection creation and dynamic (plastic) connection creation programs that run on the neuromorphic computer. The network connection driver library component is used to create connection matrix data in the spiking neural network instance of fused glial cells input by the user on the neuromorphic computer. The hardware driver library component includes an IO interface and a hardware driver. The IO interface communicates data with the neuromorphic computer, including downloading network configuration parameters and applications, uploading running process data and recorded data, etc. The hardware driver is used to implement the downloading of network configuration parameters and applications, uploading running process data and recorded data, and connecting the neuromorphic computer to the system.

[0069] The custom model creation tool is used to provide a standardized method for creating custom network node models. The standardized method for creating custom network node models is specifically based on the neural network node driver editor provided by the manufacturer, and provides a standardized editing template based on the open source PyNN interface (a universal interface for neural network simulators). The neural network nodes, i.e., nerve cells, are divided into four main parts: receptor dynamics, conduction dynamics, membrane potential dynamics, and information release. This simplifies the difficulty for users to create neural network node drivers, allowing users to perform secondary development, and making it easier for users to change and update network node models according to actual needs, so as to quickly and conveniently build BGSNN instances. Among them, the process of customizing the creation of a multi-input and multi-output network node model driver is as follows: Figure 3 As shown, since it is not the focus of protection of the present invention, it will not be described here in detail.

[0070] The neuromorphic computer driver manager is used to call the correct hardware driver from the driver library component and select and download the correct neural network driver to the neuromorphic computer based on the spiking neural network instance of the fused brain glial cells and the neuromorphic computer model, and drive the neuromorphic computer to complete the simulation of the spiking neural network instance of the fused brain glial cells. For example, based on the BGSNN instance and neuromorphic computer model input by the user, the correct hardware driver is called from the driver library and the correct neural network driver (including but not limited to network node driver and network connection driver) is selected and downloaded to the neuromorphic computer, and the neuromorphic computer completes the BGSNN simulation process. In particular, the neural network driver is different from the hardware driver. It is the driver required by the neuromorphic computer to simulate the neural network. It is generally provided by the neuromorphic computer manufacturer or manually created based on the neural network driver creation tool provided by the manufacturer. It is mainly divided into network node driver and network connection driver. The network node driver enables the neuromorphic computer to simulate the dynamic process of nerve cells, and the network connection driver enables the neuromorphic computer to simulate the dynamic process of neural network connections.

[0071] The present invention also provides a pulse neural network simulation method for fusion of brain glial cells, which is applied to the pulse neural network simulation system for fusion of brain glial cells provided above. Figure 2 As shown, the pulse neural network simulation method integrating brain glial cells includes:

[0072] Get an example of a spiking neural network that fuses glial cells.

[0073] Determine whether a neuron network node exists in the pulse neural network instance of the fused glial cells, and obtain a first determination result.

[0074] When the first determination result indicates that a neuron network node exists in the spiking neural network instance of the fused glial cells, after generating neuron model data, determining whether the spiking neural network instance of the fused glial cells exists in the glial network instance to obtain a second determination result. The neuron model data includes: pulse receptor dynamics data, dendritic conduction dynamics data, neuron membrane potential dynamics data, and neuron axonal spike firing data.

[0075] When the first determination result is that there is no neuron network node in the pulse neural network instance of the fused glial cells, it is directly determined whether there is a glial network node in the pulse neural network instance of the fused glial cells to obtain a second determination result.

[0076] When the second determination result is that there are glial network nodes in the pulse neural network instance of the fused glial cells, after generating the glial cell model data, it is determined whether there is a connection between the nerve cell populations in the pulse neural network instance of the fused glial cells to obtain the third determination result. The glial cell model data includes: neuron ion information receptor dynamics data, glial gap information receptor dynamics data, glial cell membrane potential dynamics data, glial cell glial protrusion information release data, glial cell glial gap information release data and glial cell glial ion information release data. Among them, if Figure 4 As shown, neuron model data and brain glial cell model data are used as network node application data.

[0077] When the second determination result is that there is no glial network node in the pulse neural network instance of the fused glial cells, it is directly determined whether there is a connection between the nerve cell populations in the pulse neural network instance of the fused glial cells to obtain a third determination result.

[0078] When the third determination result is that there is a connection between the nerve cell populations in the pulse neural network instance of the fused glial cells, after generating the network connection application data, the driver is called to complete the simulation of the pulse neural network instance of the fused glial cells. Figure 5 As shown, the network connection application data includes: glial gap connection data, neuron ion connection data, synaptic connection data, glial protrusion connection data and glial ion connection data.

[0079] When the third determination result is that there is no connection between the nerve cell populations in the pulse neural network instance of the fused glial cells, the driver is directly called to complete the simulation of the pulse neural network instance of the fused glial cells.

[0080] Preferably, if Figure 2 As shown in Figure 2, the process of generating neuron model data includes:

[0081] The neuron pulse receptor dynamics user configuration parameters are read, and the neuron pulse dynamics parameter generation method corresponding to the neuron pulse receptor dynamics user configuration parameters is called to generate pulse receptor dynamics data.

[0082] The neuron pulse dendritic conduction dynamics user configuration parameters are read, and the neuron pulse dendritic conduction dynamics parameter generation method corresponding to the neuron pulse dendritic conduction dynamics user configuration parameters is called to generate dendritic conduction dynamics data.

[0083] Read the neuron membrane potential dynamics user configuration parameters, and call the neuron membrane potential dynamics parameter generation method corresponding to the neuron membrane potential dynamics user configuration parameters to generate neuron membrane potential dynamics data.

[0084] The neuron axon hillock pulse emission user configuration parameters are read, and the neuron axon hillock pulse emission parameter generation method corresponding to the neuron axon hillock pulse emission user configuration parameters is called to generate the neuron axon hillock pulse emission data.

[0085] Among them, such as Figure 2 As shown in Figure 2, the process of generating brain glial cell model data includes:

[0086] Read the user-configured parameters for neuronal ion information receptor dynamics, and invoke the corresponding neuronal ion information receptor dynamics parameter generation method to generate neuronal ion information receptor dynamics data. Read the user-configured parameters for glial clearance information receptor dynamics, and invoke the corresponding glial clearance information receptor dynamics parameter generation method to generate glial clearance information receptor dynamics data. Read the user-configured parameters for glial cell membrane potential dynamics, and invoke the corresponding glial cell membrane potential dynamics parameter generation method to generate glial cell membrane potential dynamics data. Read the user-configured parameters for glial cell glial process information release, and invoke the corresponding glial cell glial process information release parameter generation method to generate glial cell glial process information release data. Read the user-configured parameters for glial cell glial clearance information release, and invoke the corresponding glial cell glial clearance information release parameter generation method to generate glial cell glial clearance information release data. Read the user configuration parameters for the issuance of brain glial cell glial ion information, and call the brain glial cell glial ion information issuance parameter generation method corresponding to the user configuration parameters for the issuance of brain glial cell glial ion information to generate brain glial cell glial ion information issuance data.

[0087] The process of generating network connection application data includes:

[0088] When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused brain glial cells is a synaptic network connection, the synaptic connection user configuration parameters are read, and the synaptic connection parameter generation method corresponding to the synaptic connection user configuration parameters is called to generate synaptic connection data.

[0089] When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused brain glial cells is a neuron ion connection, the neuron ion connection user configuration parameters are read, and the neuron ion connection parameter generation method corresponding to the neuron ion connection user configuration parameters is called to generate neuron ion connection data.

[0090] When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused brain glial cells is a glial gap junction, the glial gap connection user configuration parameters are read, and the glial gap connection parameter generation method corresponding to the glial gap connection user configuration parameters is called to generate glial gap connection data.

[0091] When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused brain glial cells is a glial process connection, the glial process connection user configuration parameters are read, and the glial process connection parameter generation method corresponding to the glial process connection user configuration parameters is called to generate glial process connection data.

[0092] When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused brain glial cells is a glial ion connection, the glial ion connection user configuration parameters are read, and the glial ion connection parameter generation method corresponding to the glial ion connection user configuration parameters is called to generate glial ion connection data.

[0093] like Figure 2 As shown in the figure, the process of calling the driver to complete the simulation of the pulse neural network example integrating brain glial cells includes:

[0094] After copying the neuron model driver in the network node driver library component, the brain glial cell model driver in the network node driver library component, or the static connection creation and dynamic connection creation programs in the network connection driver library component based on the neuron model data, brain glial cell model data, or network connection application data, the IO interface in the hardware driver library component is called to complete data interaction with the neuromorphic computer to complete the simulation of the pulse neural network instance that integrates brain glial cells.

[0095] An example is provided below to illustrate the specific implementation process of the spiking neural network simulation system and method for integrating brain glial cells provided above.

[0096] like Figure 2 As shown in the figure, the implementation process is as follows:

[0097] (1) Neuromorphic computer hardware identification and driver loading. The specific implementation method is to add drivers for different types of neuromorphic computers to the kernel. When loading the drivers, the drivers are compared with the device information one by one to determine whether the two match. If the driver matches a device, the device can be operated through the driver.

[0098] (2) Read the application of user input, that is, the BGSNN model input by the user, and extract user configuration parameters, such as the number of network nodes, network node types and parameters, the number of network connections, network connection types and parameters, etc. Existing pulse neural network development software such as Brain2 and SpikingJelly do not support the simulation of brain glial cells, while brain glial cells play an important role in the brain's advanced intelligent functions. Whether in the field of brain science or neural computing, the advanced functions of brain glial cells have become a hot topic of research. The neural cell component of this embodiment includes brain glial cells, which can support users to create brain glial cell nodes. In addition, this embodiment also provides network connections between brain glial cells and neurons, such as neuron-to-neuron ion connections, glial gap connections, glial protrusion connections from brain glial cells to synapses, and glial ion connections from brain glial cells to neurons. These connections can construct BGSNN to promote the research and simulation of the synergistic effects of glial cells and neurons, and enhance the intelligent computing potential of pulse neural networks.

[0099] (3) Generate the data required for the neural network driver, namely the neural network driver configuration data. Based on the user configuration parameters obtained in step (2) and the hardware type identified in step (1), the BGSNN manager generates the corresponding network node driver data and network connection driver data by calling the neuron dynamics component, brain glial cell dynamics component, and neural cell connection matrix component that match the hardware and meet the user configuration, such as Figure 4 and Figure 5 The specific process is as follows:

[0100] (31) According to the network node type obtained in step (2), determine whether the BGSNN model input by the user has a neuron network node type. If not, execute step (35). Otherwise, according to the receptor dynamics user configuration parameters in the network node parameters obtained in step (2), call the corresponding neuron pulse receptor dynamics parameter generation method to generate pulse receptor dynamics data.

[0101] (32) According to the neuron pulse dendritic conduction dynamics user configuration parameters in the network node parameters obtained in step (2), the corresponding neuron dendritic conduction dynamics parameter generation method is called to generate dendritic conduction dynamics data.

[0102] (33) According to the neuron membrane potential dynamics user configuration parameters obtained in step (2), the corresponding neuron membrane potential dynamics parameter generation method is called to generate neuron membrane potential dynamics data.

[0103] (34) According to the neuron axon hillock pulse emission user configuration parameters obtained in step (2), the corresponding neuron axon hillock pulse emission parameter generation method is called to generate neuron axon hillock pulse emission data.

[0104] (35) According to the network node type obtained in step (2), determine whether the BGSNN model input by the user contains a brain glial cell network node type. If not, execute step (311). Otherwise, according to the neuron ion information receptor dynamics user configuration parameters in the network node parameters obtained in step (2), call the corresponding neuron ion information receptor dynamics parameter generation method to generate neuron ion information receptor dynamics data.

[0105] (36) Based on the glial gap information receptor dynamics user configuration parameters obtained in step (2), the corresponding glial gap information receptor dynamics parameter generation method is called to generate glial gap information receptor dynamics data.

[0106] (37) According to the user configuration parameters of the brain glial cell membrane potential dynamics obtained in step (2), the corresponding brain glial cell membrane potential dynamics parameter generation method is called to generate brain glial cell membrane potential dynamics data.

[0107] (38) According to the user configuration parameters of the brain glial cell glial process information release obtained in step (2), the corresponding brain glial cell membrane potential dynamics parameter generation method is called to generate the brain glial cell membrane potential dynamics data. The corresponding brain glial cell glial process information release parameter generation method is called to generate the brain glial cell glial process information release data.

[0108] (39) According to the brain glial cell glial gap information release user configuration parameters obtained in step (2), the corresponding brain glial cell glial gap information release parameter generation method is called to generate the brain glial cell glial gap information release data.

[0109] (310) According to the brain glial cell glial ion information release user configuration parameters obtained in step (2), the corresponding brain glial cell glial ion information release parameter generation method is called to generate brain glial cell glial ion information release data.

[0110] (311) According to the network connection type obtained in step (2), determine whether there is a synaptic network connection type in the BGSNN model input by the user. If not, execute step (312). Otherwise, according to the synaptic connection user configuration parameters (including but not limited to connection weight, transmission delay, excitability or inhibition, plasticity type) in the network connection parameters obtained in step (2), call the corresponding synaptic connection parameter generation method to generate synaptic connection data.

[0111] (312) According to the network connection type obtained in step (2), determine whether the neuron ion network connection type exists in the BGSNN model input by the user. If not, jump to step (313). Otherwise, according to the neuron ion connection user configuration parameters (including but not limited to connection weight, transmission delay, excitability or inhibition, plasticity type) in the network connection parameters obtained in step (2), call the corresponding neuron ion connection parameter generation method to generate neuron ion connection data.

[0112] (313) According to the network connection type obtained in step (2), determine whether there is a glial gap network connection type in the BGSNN model input by the user. If not, jump to step (314). Otherwise, according to the glial gap connection user configuration parameters (including but not limited to connection weight, transmission delay, excitability or inhibition, plasticity type) in the network connection parameters obtained in step (2), call the corresponding glial gap connection parameter generation method to generate glial gap connection data.

[0113] (314) According to the network connection type obtained in step (2), determine whether there is a glial process network connection type in the BGSNN model input by the user. If not, jump to step (315). Otherwise, according to the glial process connection user configuration parameters (including but not limited to connection weight, transmission delay, excitability or inhibition, plasticity type) in the network connection parameters obtained in step (2), call the corresponding glial process connection parameter generation method to generate glial process connection data.

[0114] (315) According to the network connection type obtained in step (2), determine whether there is a colloid network connection type in the BGSNN model input by the user. If not, jump to step (316). Otherwise, according to the colloid connection user configuration parameters (including but not limited to connection weight, transmission delay, excitability or inhibition, plasticity type) in the network connection parameters obtained in step (2), call the corresponding colloid connection parameter generation method to generate colloid connection data.

[0115] (316) Based on the network connection type and network node type obtained in step (2), copy the matching neuron model driver in the network node driver library, and copy the matching static connection creation driver or dynamic (plastic) connection creation driver in the network connection driver library.

[0116] (4) Downloading the neural network driver and neural network driver configuration data to the neuromorphic computer. The neuromorphic computer driver manager copies the matching neuron model driver in the network node driver library and the matching static connection creation driver or dynamic (plastic) connection creation driver in the network connection driver library based on the user configuration parameters obtained in step (2) and the hardware type identified in step (1), and based on the network connection type and network node type obtained in step (2), and then calls the IO interface to download the above programs and data to the neuromorphic computer.

[0117] (5) Start the BGSNN simulation. Specifically, the IO interface is called and a “start” signal is sent (to start the simulation). After the neuromorphic computer receives the start simulation signal, it runs the neural network driver.

[0118] Based on the above, the present invention has the following beneficial effects compared with the prior art:

[0119] 1. The BGSNN implementation software framework provided by the present invention is suitable for neuromorphic computers, which enables neuromorphic computers to have the ability to construct and simulate neural networks in which brain glial cells and neurons coexist, so as to promote the application development of neuromorphic computers and enhance the intelligent computing capabilities of neuromorphic computers.

[0120] 2. The present invention provides a standardized unified hardware interface to be compatible with various types of neuromorphic computers.

[0121] 3. The present invention provides a standardized custom network node dynamics model, namely a neural network node driver creation function. It provides a standardized editing template based on the neural network node driver editor provided by the manufacturer and the open source PyNN interface (a universal interface for neural network simulators), which simplifies the difficulty of users creating neural network node drivers, so that users can quickly modify and update network node dynamics models according to their own needs.

[0122] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0123] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A pulse neural network simulation system integrating brain glial cells, characterized in that: include: Simulation software frameworks and neuromorphic computers; The simulation software framework performs data interaction with the neuromorphic computer; the simulation software framework is used to obtain the spiking neural network of the fused glial cells, and to generate a neural network driver and neural network driver configuration data based on the spiking neural network of the fused glial cells; The neuromorphic computer is used to realize the simulation of a spiking neural network integrating brain glial cells based on the neural network driver and the neural network driver configuration data; The simulation software framework includes: neuron dynamics component, brain glial cell dynamics component, neural cell connection matrix component, BGSNN component manager, driver library component, neuromorphic computer driver manager and custom model creation tool; The neuron dynamics component, the brain glial cell dynamics component and the neural cell connection matrix component all interact with the BGSNN component manager for data; the neuromorphic computer driver manager is connected to the neuron dynamics component, the brain glial cell dynamics component and the neural cell connection matrix component respectively; the neuromorphic computer driver manager interacts with the driver library component for data; the driver library component interacts with the neuromorphic computer for data; the driver library component is connected to the custom model creation tool; The neuron dynamics component is embedded with neuron pulse receptor dynamics, neuron pulse dendrite conduction dynamics, neuron membrane potential update program and neuron axon pulse emission program, so as to process and respond to the transmitted neuron pulse signals in the time domain; The brain glial cell dynamics component is implanted with neuronal ion information receptor dynamics, glial gap information receptor dynamics, neuronal ion conduction dynamics, glial gap conduction dynamics, brain glial cell membrane potential update program, brain glial cell glial protrusion information release program, brain glial cell glial gap information release program and brain glial cell glial ion information release program, so as to process and respond to the conducted neuronal ion information and glial gap information in the time domain; The neural cell connection matrix component is used to construct connections between neural cell populations in a spiking neural network; The BGSNN component manager is used to obtain and parse the pulse neural network instance of the fused brain glial cells, and to select corresponding implementation modules from the neuron dynamics component, the brain glial cell dynamics component, and the neural cell connection matrix component to generate network configuration parameters and network initial state parameters corresponding to the pulse neural network instance of the fused brain glial cells; The driver library component is used to simulate the nodes of the pulse neural network instance of the fused brain glial cells, to create the connection matrix data of the pulse neural network instance of the fused brain glial cells, and to realize data communication with the neuromorphic computer; The custom model creation tool is used to provide a standardized method for creating a custom network node model; The neuromorphic computer driver manager is used to call the correct hardware driver from the driver library component and select and download the correct neural network driver to the neuromorphic computer according to the pulse neural network instance of the fused brain glial cells and the neuromorphic computer model, and drive the neuromorphic computer to complete the simulation of the pulse neural network instance of the fused brain glial cells.

2. The spiking neural network simulation system integrating brain glial cells according to claim 1 is characterized in that: The driver library components include: a network node driver library component, a network connection driver library component and a hardware driver library component; The network node driver library component is connected to the custom model creation tool; The network node driver library component includes a neuron model driver and a glial cell model driver running on the neuromorphic computer; the network node driver library component is used to simulate a node of a pulse neural network instance of fused glial cells input by a user on the neuromorphic computer; The network connection driver library component includes static connection creation and dynamic connection creation programs running on the neuromorphic computer; the network connection driver library component is used to create connection matrix data in the pulse neural network instance of the fused glial cells input by the user on the neuromorphic computer; The hardware driver library component includes an IO interface and a hardware driver; the IO interface communicates data with the neuromorphic computer; the hardware driver is used to implement the downloading of network configuration parameters and applications and the uploading of running process data and recorded data.

3. The pulse neural network simulation system integrating brain glial cells according to claim 1 is characterized in that: The connections between neural cell populations in spiking neural networks include synaptic connections, neuronal ionic connections, glial gap junctions, glial process connections, and glial ionic connections.

4. The spiking neural network simulation system integrating brain glial cells according to claim 1 is characterized in that: The standardized method for creating a customized network node model is based on a neural network node driver editor provided by the manufacturer, and provides a standardized editing template based on the open source PyNN interface, which divides the neural network nodes into four parts: receptor dynamics, conduction dynamics, membrane potential dynamics, and information release.

5. A pulse neural network simulation method integrating brain glial cells, characterized in that: Applicable to the pulse neural network simulation system integrating brain glial cells as claimed in any one of claims 1 to 4; The pulse neural network simulation method for integrating brain glial cells includes: Obtain an example of a spiking neural network that fuses brain glial cells; Determining whether there is a neuron network node in the pulse neural network instance of the fused glial cells to obtain a first determination result; When the first determination result is that a neuron network node exists in the pulse neural network instance of the fused glial cells, after generating neuron model data, determining whether a glial network node exists in the pulse neural network instance of the fused glial cells to obtain a second determination result; the neuron model data includes: pulse receptor dynamics data, dendritic conduction dynamics data, neuron membrane potential dynamics data, and neuron axon pulse discharge data; When the first determination result is that there is no neuron network node in the pulse neural network instance of the fused glial cells, directly determining whether there is a glial network node in the pulse neural network instance of the fused glial cells to obtain a second determination result; When the second determination result is that a glial network node exists in the pulse neural network instance of the fused glial cells, after generating the glial cell model data, determining whether there is a connection between the neural cell populations in the pulse neural network instance of the fused glial cells, and obtaining a third determination result; the glial cell model data includes: neuron ion information receptor dynamics data, glial gap information receptor dynamics data, glial cell membrane potential dynamics data, glial cell glial process information release data, glial cell glial gap information release data, and glial cell glial ion information release data; When the second determination result is that there is no glial network node in the spiking neural network instance of the fused glial cells, directly determining whether there is a connection between the nerve cell populations in the spiking neural network instance of the fused glial cells to obtain a third determination result; When the third determination result indicates that there is a connection between the nerve cell populations in the spiking neural network instance of the fused glial cells, after generating network connection application data, calling a driver to complete the simulation of the spiking neural network instance of the fused glial cells; the network connection application data includes: glial gap connection data, neuron ion connection data, synaptic connection data, glial process connection data, and glial ion connection data; When the third determination result is that there is no connection between the nerve cell populations in the pulse neural network instance of the fused glial cells, the driver is directly called to complete the simulation of the pulse neural network instance of the fused glial cells.

6. The method for simulating a spiking neural network integrating brain glial cells according to claim 5, characterized in that: The process of generating neuron model data includes: Reading a neuron pulse receptor dynamics user configuration parameter, and calling a neuron pulse dynamics parameter generation method corresponding to the neuron pulse receptor dynamics user configuration parameter to generate pulse receptor dynamics data; Reading a neuron pulse dendrite conduction dynamics user configuration parameter, and calling a neuron pulse dendrite conduction dynamics parameter generation method corresponding to the neuron pulse dendrite conduction dynamics user configuration parameter to generate dendrite conduction dynamics data; Reading neuronal membrane potential dynamics user configuration parameters, calling a neuronal membrane potential dynamics parameter generation method corresponding to the neuronal membrane potential dynamics user configuration parameters to generate neuronal membrane potential dynamics data; The neuron axon hillock pulse emission user configuration parameters are read, and a neuron axon hillock pulse emission parameter generation method corresponding to the neuron axon hillock pulse emission user configuration parameters is called to generate neuron axon hillock pulse emission data.

7. The method for simulating a spiking neural network integrating brain glial cells according to claim 5, wherein: The process of generating brain glial cell model data includes: Read the neuron ion information receptor dynamics user configuration parameters, call the neuron ion information receptor dynamics parameter generation method corresponding to the neuron ion information receptor dynamics user configuration parameters to generate neuron ion information receptor dynamics data; read the glial gap information receptor dynamics user configuration parameters, call the glial gap information receptor dynamics parameter generation method corresponding to the glial gap information receptor dynamics user configuration parameters to generate glial gap information receptor dynamics data; read the brain glial cell membrane potential dynamics user configuration parameters, call the brain glial cell membrane potential dynamics parameter generation method corresponding to the brain glial cell membrane potential user configuration parameters to generate brain glial cell membrane potential dynamics data; read the glial processes of the brain glial cells The information is released by the user configuration parameters, and the method for generating the glial cell glial protrusion information release parameters corresponding to the glial cell glial protrusion information release user configuration parameters is called to generate the glial cell glial protrusion information release data; the glial cell glial gap information release user configuration parameters are read, and the method for generating the glial cell glial gap information release parameters corresponding to the glial cell glial gap information release user configuration parameters is called to generate the glial cell glial gap information release data; the glial cell glial ion information release user configuration parameters are read, and the method for generating the glial cell glial ion information release parameters corresponding to the glial cell glial ion information release user configuration parameters is called to generate the glial cell glial ion information release data.

8. The method for simulating a spiking neural network integrating brain glial cells according to claim 5, wherein: The process of generating network connection application data includes: When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused glial cells is a synaptic network connection, reading the synaptic connection user configuration parameters, and calling the synaptic connection parameter generation method corresponding to the synaptic connection user configuration parameters to generate synaptic connection data; When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused glial cells is a neuron ion connection, reading the neuron ion connection user configuration parameters, and calling the neuron ion connection parameter generation method corresponding to the neuron ion connection user configuration parameters to generate neuron ion connection data; When the third determination result is that the connection between the nerve cell populations in the spiking neural network instance of the fused brain glial cells is a glial gap junction, reading the glial gap junction user configuration parameters, and calling the glial gap junction parameter generation method corresponding to the glial gap junction user configuration parameters to generate glial gap junction data; When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused brain glial cells is a glial process connection, reading the glial process connection user configuration parameters, and calling the glial process connection parameter generation method corresponding to the glial process connection user configuration parameters to generate glial process connection data; When the third determination result is that the connection between the nerve cell populations in the pulse neural network instance of the fused brain glial cells is a glial ion connection, the glial ion connection user configuration parameters are read, and the glial ion connection parameter generation method corresponding to the glial ion connection user configuration parameters is called to generate glial ion connection data.

9. The method for simulating a spiking neural network integrating brain glial cells according to claim 5, wherein: The process of calling the driver to complete the simulation of the spiking neural network example integrating brain glial cells includes: After correspondingly copying the neuron model driver in the network node driver library component, the brain glial cell model driver in the network node driver library component, or the static connection creation and dynamic connection creation programs in the network connection driver library component based on the neuron model data, the brain glial cell model data, or the network connection application data, the IO interface in the hardware driver library component is called to complete data interaction with the neuromorphic computer to complete the simulation of the pulse neural network instance that integrates brain glial cells.

Images