Brain-like information transient transmission method and device based on neural network

Abstract

The invention provides a brain-like information transient transmission method and device based on a neural network, and the method comprises the steps: obtaining different types of input data and test data corresponding to the input data; simulating a transient transmission process of coded information in the brain through an information transient transmission model based on the input data to obtain a fixed transmission path for dynamic distribution of neurons; after a preset target time, inputting the test data into the information transient transmission model, and obtaining a matching result output by the information transient transmission model through a fixed transmission path; wherein the information transient transfer model is constructed on the basis of a recurrent neural network, the output of any neuron in a hidden layer of the information transient transfer model is determined on the basis of the output of other neurons at a target moment, and the target moment is determined on the basis of the external input and the inhibition intensity of any neuron; according to the invention, the interpretability and biological rationality of the neural network can be improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for transient transmission of brain-like information based on a neural network. Background Art

[0002] Artificial Neural Networks (ANNs) process and transmit information through a set of interconnected nodes (i.e., artificial neurons). The connections between these nodes have different weights, which determine the degree to which the input signal affects the output of the neuron. Each neuron receives input signals from other neurons, performs a weighted sum of these signals, and generates an output signal through an activation function.

[0003] In today's artificial neural networks, the model's memory process is often static, and once the information is stored in the memory unit, it will no longer change the neuron's encoding of the information. This memory of information will lead to the solidification of the neuron's response to the information. Neurons need to maintain the firing state to encode information, and often lack the flexibility of information encoding and the dynamic process of information transmission. On the contrary, the encoding of information by neurons in the human brain is often dynamic. After encoding and transmitting information, neurons will return to an unexcited state, achieving a dynamic effect of firing.

[0004] This shows that the brain-like information transient transmission method based on neural network in related technologies has the technical problem of lack of biological rationality. Summary of the invention

[0005] The present invention provides a method and device for transient transmission of brain-like information based on a neural network, so as to solve the defect of the transient transmission method of brain-like information based on a neural network in the prior art that the method lacks biological rationality, and realizes a process that can more reasonably simulate the transient encoding of information in the brain.

[0006] The present invention provides a method for transient information transmission of brain-like based on neural network, comprising the following steps. Input data of different categories and test data corresponding to the input data are obtained; based on the input data, a transient information transmission model is used to simulate the process of transient transmission of coded information in the brain to obtain a fixed transmission path for dynamic neuron emission; after a preset target time, the test data is input into the transient information transmission model and obtained through the fixed transmission path to obtain a matching result output by the transient information transmission model, wherein the matching result is used to indicate the category matching between the test data and the input data; wherein the transient information transmission model is constructed based on a recurrent neural network, and the output of any neuron in the hidden layer of the transient information transmission model is determined based on the output of other neurons at the target time, and the target time is determined based on the external input and the inhibition strength of any neuron.

[0007] According to a neural network-based brain-like information transient transmission method provided by the present invention, the information transient transmission model includes: an input layer, a hidden layer and an output layer; wherein the hidden layer includes: sensory brain areas, cognitive brain areas and decision-making brain areas, and the sensory brain areas, the cognitive brain areas and the decision-making brain areas are hierarchically connected, and the hierarchical connection is used to simulate cross-brain area information transmission in the brain.

[0008] According to a neural network-based brain-like information transient transmission method provided by the present invention, the neurons in the hidden layer of the information transient transmission model include: excitatory neurons and inhibitory neurons; wherein the excitatory neurons are used to simulate the excitatory synapses of the brain, and the inhibitory neurons are used to simulate the inhibitory synapses of the brain.

[0009] According to a method for transient transmission of brain-like information based on a neural network provided by the present invention, the method also includes: determining the firing rate of the target neuron at the current moment according to the firing state of the target neuron at the previous moment, the firing states of other neurons at the previous moment, the external input received by the target neuron, and the inhibition strength of the target neuron at the current moment.

[0010] According to a method for transient transmission of brain-like information based on a neural network provided by the present invention, before determining the firing rate of the target neuron at the current moment, the method also includes: determining the inhibition strength of the target neuron at the current moment based on the inhibition strength of the target neuron at the previous moment and the firing state of the target neuron at the current moment.

[0011] According to a neural network-based brain-like information transient transmission method provided by the present invention, before the information transient transmission model simulates the process of transient transmission of encoded information in the brain based on the input data to obtain a fixed transmission path for dynamic neuron emission, the method also includes: obtaining input data samples of different categories, and test data samples corresponding to the input data samples; inputting the input data samples and the test data samples into a preset information transient transmission model respectively to obtain a training matching result output by the preset information transient transmission model; determining the cross entropy between the training matching result and the preset correct matching result, so as to use the cross entropy as a loss function; performing back propagation based on the loss function to update the preset information transient transmission model to obtain a trained information transient transmission model.

[0012] The present invention also provides a brain-like information transient transmission device based on a neural network, comprising the following modules: an acquisition module, used to acquire input data of different categories, and test data corresponding to the input data; a simulation module, used to simulate the process of transient transmission of encoded information in the brain based on the input data through an information transient transmission model, and obtain a fixed transmission path for dynamic neuron emission; an output module, used to input the test data into the information transient transmission model after a preset target time and obtain a matching result output by the information transient transmission model through the fixed transmission path, wherein the matching result is used to indicate the category matching between the test data and the input data; wherein the information transient transmission model is constructed based on a recurrent neural network, and the output of any neuron in the hidden layer of the information transient transmission model is determined based on the output of other neurons at the target moment, and the target moment is determined based on the external input and the inhibition strength of any neuron.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for transient transmission of brain-like information based on a neural network as described above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for transient transmission of brain-like information based on a neural network as described above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned methods for transient transmission of brain-like information based on a neural network.

[0016] The present invention provides a method and device for transient information transmission based on a neural network, which simulates the process of transient transmission of encoded information in the brain based on different categories of input data through an information transient transmission model constructed based on a recurrent neural network, and obtains a fixed transmission path for dynamic neuron release. Thus, the model simulates the process of transient transmission and memory of encoded information in the brain to form a transmission path. After a preset target time, test data corresponding to the input data is input into the information transient transmission model to obtain a category matching result between the test data output by the information transient transmission model and the input data, wherein the output of any neuron in the hidden layer of the information transient transmission model is determined based on the output of other neurons at the target moment, and the target moment is determined based on the external input and the inhibition strength of any neuron. Thus, the process of transient encoding of information in the brain can be simulated more reasonably, and energy consumption can be reduced while maintaining performance. This solves the technical problem of the lack of biological rationality in the method for transient information transmission based on a neural network in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a schematic flow chart of the method for transient transmission of brain-like information based on neural network provided by the present invention.

[0019] Figure 2 It is a structural schematic diagram of the information transient transmission model provided by the present invention.

[0020] Figure 3 It is a schematic diagram of the structure of the neural network-based brain-like information transient transmission device provided by the present invention.

[0021] Figure 4 It is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0023] In recent years, many different types of artificial deep neural networks have been proposed to solve various types of classification, recognition and prediction problems. However, while artificial deep neural networks are developing rapidly, there are also some shortcomings. The first problem is that the training process of the network often lacks a reasonable explanation. The complex internal connection structure of the deep neural network makes it impossible for people to know the role of the specific internal structure. The second problem is the increase in artificial neural network parameters. The effect of the neural network often depends on the adjustment of a large number of network parameters, which has high labor and time costs. The third problem is that many techniques in deep neural networks are often considered purely from the perspective of improving the model effect, and are not biologically reasonable. Therefore, in recent years, artificial neural networks hope to get help from brain science, optimize artificial neural networks by studying the working methods of neurons in the brain, and improve the interpretability and biological rationality of neural networks.

[0024] When processing time series information, a recurrent neural network (RNN) can be used. Compared with a deep neural network, a recurrent neural network has a loop structure that retains its own output as input, and can better act on data at different times. In the human brain, the firing state of neurons is often related to their own historical firing state, so recurrent neural networks can simulate the feedback process of neuronal information very well. Today's artificial neural networks often rarely consider the biological rationality of the network. Adding biological computing rules to today's artificial neural networks can often improve the biological rationality of the network and reduce the computing energy consumption of the network.

[0025] In today's artificial neural networks, the model's memory process is often static, and once the information is stored in the memory unit, it will no longer change the neuron's encoding of the information. This memory of information will lead to the solidification of the neuron's response to the information. Neurons need to maintain the firing state to encode information, and often lack the flexibility of information encoding and the dynamic process of information transmission. On the contrary, the encoding of information by neurons in the human brain is often dynamic. After encoding and transmitting information, neurons will return to an unexcited state, achieving a dynamic effect of firing.

[0026] The present invention provides a method for transient information transmission based on a neural network, which is used to explain the way in which information is transiently transmitted in the brain. It also improves the biological rationality of artificial neural networks, reduces the learning energy consumption of the network, and has great advantages for later efficient on-chip computing.

[0027] The method provided by the embodiment of the present invention is applicable to the problem of data matching and data classification in the presence of delay. Taking the data matching problem as an example, different data codes are input, and the corresponding matching codes are input after a period of time. The model will output whether the two codes match, and the neurons in the process of input data transmission in the model are dynamically fired. Different input neurons are fired at different times, which is explained below.

[0028] refer to Figure 1 , Figure 1 is a flow chart of the method for transient transmission of brain-like information based on a neural network provided by the present invention, such as Figure 1 As shown, the method includes the following: Step 101, obtaining input data of different categories and test data corresponding to the input data.

[0029] In the embodiment of the present invention, because the subsequent information transient transfer model is applicable to the data matching problem, the input data must have the following characteristics, and each input data has a corresponding matching result. Since the information transient transfer model uses a recurrent neural network, the input data needs to conform to the time step of the recurrent neural network. At the same time, since the information transient transfer model is used to simulate the process of information transfer, the data needs to be input only in a fixed time step and no input in the remaining time steps.

[0030] Step 102, using the information transient transmission model to simulate the process of transient transmission of encoded information in the brain based on input data, and obtain a fixed transmission path for dynamic firing of neurons.

[0031] In an embodiment of the present invention, different categories of input data are encoded and input into a preset information transient transmission model; the information transient transmission model simulates the process of transient transmission of encoded information in the brain, forming a fixed transmission path for dynamic neuronal firing.

[0032] Step 103, after a preset target time, input the test data into the information transient transfer model and obtain the matching result output by the information transient transfer model through a fixed transfer path, wherein the matching result is used to indicate the category matching between the test data and the input data.

[0033] Among them, the information transient transmission model is constructed based on a recurrent neural network. The output of any neuron in the hidden layer of the information transient transmission model is determined based on the outputs of other neurons at the target moment, and the target moment is determined based on the external input and the inhibition strength of any neuron.

[0034] In an embodiment of the present invention, different categories of input data and matching data corresponding to the input data are generated; different categories of input data are encoded and input into the information transient transfer model; the information transient transfer model simulates the process of transient transmission of encoded information in the brain, forming a fixed transmission path for dynamic release of neurons; after a period of time, different types of matching data are input into the information transient transfer model, and the information transient transfer model determines whether the input data matches the matching data; wherein, the information transient transfer model is constructed based on a recurrent neural network, and the output of any neuron in the hidden layer of the information transient transfer model is based on the output of other neurons at the previous moment, the external input at that moment, and the inhibition strength of the neuron itself. Hierarchical connections between brain regions are used to simulate information transmission across brain regions; excitatory neurons and inhibitory neurons are used to simulate excitatory and inhibitory synapses in the brain.

[0035] In an embodiment of the present invention, an information transient transmission model is constructed based on a recurrent neural network, and the firing state of any neuron in the hidden layer of the information transient transmission model is determined based on the firing state of the neuron at the previous moment and the firing states of other neurons at the current moment, and the firing state of the neuron is also related to its own inhibition strength; the neuron's own inhibition strength is determined based on the neuron's own inhibition strength at the previous moment and the firing state of the neuron at the current moment; neurons in the same hidden layer are sparsely connected, and neurons are divided into different brain regions to simulate sensory brain regions, cognitive brain regions and decision-making brain regions in the brain, and different connection probability weights and distance weights are set between different brain regions and in the same brain region.

[0036] Specifically, the recognition result is the matching result between the input information and the test information. For example, if the code of a specific information is input, and the code matching the information is input during the test, the model output result is that the information matches. On the contrary, if the code that does not match the information is input during the test, the model output result is that the information does not match.

[0037] The information transient transmission model is obtained through training, and the specific training method is as follows: first, generate a data set that needs to be matched, including the encoding of the input data and the encoding of the matching data. Then, train the model based on the generated data so that the model can match and judge the generated data.

[0038] Through the above steps of the embodiment of the present invention, input data of different categories and test data corresponding to the input data are obtained; the information transient transfer model is used to simulate the process of transient transmission of encoded information in the brain based on the input data, and a fixed transmission path for dynamic neuron release is obtained; after a preset target time, the test data is input into the information transient transfer model and a matching result output by the information transient transfer model is obtained through the fixed transmission path, wherein the matching result is used to indicate the category matching between the test data and the input data; wherein the information transient transfer model is constructed based on a recurrent neural network, and the output of any neuron in the hidden layer of the information transient transfer model is determined based on the output of other neurons at the target moment, and the target moment is determined based on the external input and the inhibition strength of any neuron, thereby being able to more reasonably simulate the process of transient encoding of information in the brain, while maintaining performance and reducing energy consumption; thereby solving the technical problem of the lack of biological rationality in the brain-like information transient transmission method based on neural networks in the related technology.

[0039] According to a neural network-based brain-like information transient transmission method provided by the present invention, the information transient transmission model includes: an input layer, a hidden layer and an output layer; wherein the hidden layer includes: sensory brain areas, cognitive brain areas and decision-making brain areas, and the sensory brain areas, cognitive brain areas and decision-making brain areas are hierarchically connected, and the hierarchical connection is used to simulate cross-brain area information transmission in the brain.

[0040] The initial model of the information transient transmission model can be a recurrent neural network. The network structure of the recurrent neural network includes an input layer, an output layer and a hidden layer. The number of neurons in the input layer can be determined according to the dimension of the input data encoding; in the embodiment of the present invention, the input layer in the information transient transmission model has 24 neurons. The output layer has three neurons, representing matching, mismatching and waiting respectively.

[0041] According to a method for transient brain-like information transmission based on a neural network provided by the present invention, neurons in a hidden layer of the transient information transmission model include: excitatory neurons and inhibitory neurons; Among them, excitatory neurons are used to simulate the excitatory synapses of the brain, and inhibitory neurons are used to simulate the inhibitory synapses of the brain.

[0042] The number of neurons in the hidden layer can be designed according to actual needs. The commonly used hidden layer has 1000 neurons, including 800 excitatory neurons and 200 inhibitory neurons.

[0043] After the hidden layer is cyclically connected, the neurons in the same hidden layer can be divided to simulate different brain regions. The division of brain regions can be determined according to actual conditions, and is usually divided into three brain regions, including sensory brain regions, cognitive brain regions, and decision-making brain regions. Sparse connections are used between brain regions, and dense connections are used within brain regions, so that when the information contained in the neurons in the current hidden layer is input to the next hidden layer, the neurons in the next hidden layer can obtain richer information, which improves the model's anti-noise performance, thereby improving the robustness of the entire information transient transmission model. At the same time, the use of sparse connections can reduce the energy consumption of the model while maintaining performance.

[0044] The sparse connection method can be a random connection method. For example, the current hidden layer contains 4 neurons, namely neuron 1, neuron 2, neuron 3 and neuron 4. The outputs of neurons 1 and 2 can be connected, and the outputs of neurons 3 and 4 can be connected.

[0045] The output of the hidden layer in the recurrent neural network is composed of the firing states of the neurons in the hidden layer. The firing state of a neuron is determined by the state of the neuron at the previous moment, the states of other neurons at the current moment, and the neuron's own inhibition strength.

[0046] In the embodiment of the present invention, since the information transient transmission model is constructed with the recurrent neural network as the initial model, the neurons in the hidden layer include a forward channel and a recurrent channel. Among them, only the neurons in the sensory brain area receive input from the forward channel, and the neurons in the cognitive brain area and the decision-making brain area do not receive input from the forward channel.

[0047] The forward pass is used to connect neurons in the sensory brain area with neurons in the input layer. The product of the output of the input layer neurons in the forward pass and the input weight is used as the input of the sensory brain area neurons in the hidden layer.

[0048] The recurrent channel is used to connect the output of any neuron at the previous moment with the outputs of other neurons in the same layer at the current moment.

[0049] In an embodiment of the present invention, a transient information transfer model is constructed by a recurrent neural network, and the firing state of neurons in the sensory brain area of ​​the hidden layer in the transient information transfer model is determined based on the encoding of the input data in the forward channel, the firing state of neurons in the cyclic channel at the previous moment, and the inhibitory strength of the neurons themselves. The firing state of neurons in the cognitive brain area and the decision-making brain area in the hidden layer is determined based on the firing state of neurons in the cyclic channel at the previous moment and the inhibitory strength of the neurons themselves. Sparse connections are used between neurons in different brain areas in the same hidden layer. Since the firing state of neurons can produce specific dynamic changes according to the input in real time, it can adapt to changes in data encoding, has a higher dynamic calculation ability, and improves the robustness of the model. In addition, by adopting sparse connections, the resource overhead of model learning is reduced to a certain extent without changing the performance of the model, the memory usage is reduced, and the energy consumption is reduced, which is more in line with the simulation of neurons in the human brain.

[0050] According to a method for transient transmission of brain-like information based on a neural network provided by the present invention, the method further includes: The firing rate of the target neuron at the current moment is determined according to the firing state of the target neuron at the previous moment, the firing states of other neurons at the previous moment, the external input received by the target neuron, and the inhibition strength of the target neuron at the current moment.

[0051] Based on the above embodiment, the connection weights between the input data and the neurons in the sensory brain area in the forward channel of the information transient transfer model and between the neurons at the previous moment and the neurons at the current moment in the recurrent channel are initialized based on the gamma distribution.

[0052] The firing state of hidden layer neurons in the information transient transfer model is updated based on the following formula:

[0053] in, is the firing rate of the (target) neuron at time t, and its magnitude is determined by the firing state of the neuron at the previous moment , the neuron receives external input , the firing state of other neurons at the previous moment and the inhibitory strength of the neuron itself at the current moment Joint decision. Represents that any neuron is affected by itself at the previous moment, is the weight between the input data and the hidden layer sensory neurons, is the connection weight between any neuron in the hidden layer at time t and all neurons in the hidden layer at the previous time, The neurons themselves The suppression strength at a moment depends on the suppression strength at the previous moment. and the current state of its own release , as shown above. Where m is the intensity of the current moment The current neuron's own inhibition strength The parameters of the influence degree are updated along with the network. is Gaussian white noise.

[0054] Specifically, due to the use of brain region division and sparse connections in the transient information transmission model, the connection weights between neurons in the forward channel and input data and the connection weights between neurons in the recurrent channel will affect the firing state of neurons.

[0055] According to a method for transient brain-like information transmission based on a neural network provided by the present invention, before determining the firing rate of the target neuron at the current moment, the method further includes: The inhibition strength of the target neuron at the previous moment is determined according to the inhibition strength of the target neuron at the current moment and the firing state of the target neuron at the current moment.

[0056] In this embodiment of the present invention, the inhibitory strength of any neuron itself is updated based on the following formula:

[0057] The neurons themselves The suppression strength at a certain moment depends on the suppression strength at the previous moment. and the current state of its own release .in m To determine the intensity of the current moment The strength of neuronal inhibition v The parameters of the influence degree are updated along with the network.

[0058] In the embodiment of the present invention, the state of the neuron will be inhibited by itself, and the overall trend of dynamic emission is in line with the dynamic characteristics and biological rationality. The emission state of the neuron at this moment is determined based on the neuron's own emission state and the inhibition strength of the previous moment. In this way, the neuron is only emitted part of the time when transmitting information, and the overall state is dynamic.

[0059] In an embodiment of the present invention, the outputs of neurons in the hidden layer of the information transient transfer model are connected based on the following steps: Determine the sparse connection ratio; the sparse connection ratio is the ratio of the number of neurons connected in any hidden layer to all neurons in any hidden layer; the sparse connection ratios of neurons between different brain regions and within the same brain region are different. From the sensory brain region to the cognitive brain region and then to the decision-making brain region, it is a forward connection, and the sparse connection ratio is small; the reverse is a reverse connection, and the sparse connection ratio is large. The sparse connection ratio within the brain region is fixed.

[0060] Select neurons that meet the sparse connection ratio from any hidden layer and randomly connect the outputs of each neuron.

[0061] Specifically, when sparsely connecting neurons in the same hidden layer, a sparse connection ratio may be set to indicate the degree of sparse connection.

[0062] The sparse connection ratio is the ratio of the number of neurons connected in any hidden layer to the number of all neurons in the hidden layer. For example, when the sparse connection ratio is 60%, 60% of the neurons can be selected from all neurons in the hidden layer for sparse connection. The specific connection method is random connection.

[0063] Through the embodiments of the present invention, based on random sparse connections, the model operates in a manner more similar to the human brain, thereby improving the biological rationality of the model and reducing the energy consumption of the model.

[0064] According to a method for transient information transmission based on a neural network provided by the present invention, before simulating the process of transient transmission of coded information in the brain based on input data through an information transient transmission model to obtain a fixed transmission path for dynamic neuron emission, the method further includes: Obtain input data samples of different categories and test data samples corresponding to the input data samples; Inputting the input data sample and the test data sample into the preset information transient transfer model respectively, and obtaining the training matching result output by the preset information transient transfer model; Determine the cross entropy between the training matching result and the preset correct matching result, so as to use the cross entropy as a loss function; Back propagation is performed based on the loss function to update the preset information transient transfer model to obtain the trained information transient transfer model.

[0065] In an embodiment of the present invention, the simulation information transient transmission model is trained based on the following steps: generating input data of different categories and matching data corresponding to the input data; then, training the model according to the generated data so that the model can perform matching judgment on the generated data; passing the data set into the model input layer for training to obtain the error between the input data and the matching data, and the loss function is the cross entropy between the matching result and the correct matching result; based on backpropagation, the error information is propagated to the hidden layer neurons to realize the training of the model.

[0066] refer to Figure 2 , Figure 2 It is a structural schematic diagram of the information transient transmission model provided by the present invention, which includes an input layer, a hidden layer and an output layer, and the hidden layer includes a sensory brain area, a cognitive brain area and a decision-making brain area.

[0067] According to a method for transient transmission of brain-like information based on a neural network provided by the present invention, the method specifically includes the following steps.

[0068] Step 1: Generate input data and the matching corresponding to each input.

[0069] Step 2: Construct a transient information transfer model (recurrent spiking neural network) with sparse connections and brain region concepts.

[0070] like Figure 2 As shown, the network includes an input layer, a hidden layer, and an output layer; Figure 2 The dashed lines are sparse connections.

[0071] Step 3: Input the encoding of the input data and input the corresponding matching after a period of time. Get the matching error after the output layer.

[0072] Step 4: Calculate the error using cross entropy loss and update the neural network parameters by back-propagating the error information.

[0073] The information transient transmission method provided by the embodiment of the present invention adopts a recurrent pulse neural network as an initial model, which has the following advantages: Dynamic calculation: The addition of dynamic neurons with self-inhibition characteristics enriches the plasticity of neurons, allowing multiple neurons in the network to have specific dynamic changes, while improving the overall dynamic calculation capabilities of the network.

[0074] Low power consumption: Sparse connections between neurons can reduce computing overhead and lower power consumption without affecting performance, which is a great advantage for chip use.

[0075] Robust computation: Adjustable recurrent connections are set inside the hidden layers to help match performance, especially for noisy samples, which is more conducive to preserving timing information and robust classification.

[0076] Biological rationality: The neurons in the hidden layer of the recurrent neural network are divided into three brain regions to simulate different types of neurons in the brain. This is more in line with the division of labor of neurons in the human brain and has biological rationality.

[0077] The following is a description of the brain-like information transient transmission device based on a neural network provided by the present invention. The brain-like information transient transmission device based on a neural network described below and the brain-like information transient transmission method based on a neural network described above can be referenced to each other.

[0078] refer to Figure 3 , Figure 3 It is a structural schematic diagram of the neural network-based brain-like information transient transmission device provided by the present invention, which includes an acquisition module 301, a simulation module 302 and an output module 303.

[0079] An acquisition module 301 is used to acquire input data of different categories and test data corresponding to the input data; A simulation module 302 is used to simulate the process of transient transmission of coded information in the brain based on the input data through an information transient transmission model to obtain a fixed transmission path of dynamic neuron firing; The output module 303 is used to input the test data into the information transient transfer model and obtain the matching result of the information transient transfer model output through the fixed transmission path after a preset target time, wherein the matching result is used to indicate the category matching between the test data and the input data; wherein the information transient transfer model is constructed based on a recurrent neural network, and the output of any neuron in the hidden layer of the information transient transfer model is determined based on the outputs of other neurons at the target time, and the target time is determined based on the external input and the inhibition strength of any neuron.

[0080] Specifically, the above-mentioned neural network-based brain-like information transient transmission device provided by the present invention can implement all the method steps implemented by the above-mentioned neural network-based brain-like information transient transmission method embodiment, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.

[0081] Figure 4 is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as Figure 4As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430 and a communication bus 440, wherein the processor 410, the communication interface 420 and the memory 430 communicate with each other through the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute the brain-like information transient transmission method based on the neural network, the method comprising: obtaining input data of different categories and test data corresponding to the input data; simulating the process of transient transmission of coded information in the brain based on the input data through the information transient transmission model to obtain a fixed transmission path for dynamic neuron emission; after a preset target time, inputting the test data into the information transient transmission model and obtaining a matching result output by the information transient transmission model through the fixed transmission path, wherein the matching result is used to indicate the category matching between the test data and the input data; wherein the information transient transmission model is constructed based on a recurrent neural network, the output of any neuron in the hidden layer of the information transient transmission model is determined based on the output of other neurons at the target time, and the target time is determined based on the external input and the inhibition strength of any neuron.

[0082] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0083] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the neural network-based brain-like information transient transmission method provided by the above methods, the method including: obtaining input data of different categories, and test data corresponding to the input data; simulating the process of transient transmission of encoded information in the brain based on the input data through an information transient transmission model to obtain a fixed transmission path for dynamic neuron release; after a preset target time, inputting the test data into the information transient transmission model and obtaining a matching result output by the information transient transmission model through the fixed transmission path, wherein the matching result is used to indicate the category matching between the test data and the input data; wherein the information transient transmission model is constructed based on a recurrent neural network, and the output of any neuron in the hidden layer of the information transient transmission model is determined based on the output of other neurons at the target moment, and the target moment is determined based on the external input and the inhibition strength of any neuron.

[0084] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the neural network-based brain-like information transient transmission method provided by the above-mentioned methods, the method comprising: obtaining input data of different categories, and test data corresponding to the input data; simulating the process of transient transmission of encoded information in the brain based on the input data through an information transient transmission model, and obtaining a fixed transmission path for dynamic neuron release; after a preset target time, inputting the test data into the information transient transmission model and obtaining a matching result output by the information transient transmission model through the fixed transmission path, wherein the matching result is used to indicate the category matching between the test data and the input data; wherein the information transient transmission model is constructed based on a recurrent neural network, and the output of any neuron in the hidden layer of the information transient transmission model is determined based on the output of other neurons at the target moment, and the target moment is determined based on the external input and the inhibition strength of any neuron.

[0085] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0086] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for transient transmission of brain-like information based on a neural network, characterized in that: include: Acquire input data of different categories and test data corresponding to the input data; Based on the input data, the information transient transmission model is used to simulate the process of transient transmission of encoded information in the brain, and a fixed transmission path of dynamic neuron firing is obtained; After a preset target time, the test data is input into the information transient transmission model and passes through the fixed transmission path to obtain a matching result output by the information transient transmission model, wherein the matching result is used to indicate a category matching situation between the test data and the input data; Among them, the information transient transfer model is constructed based on a recurrent neural network, and the output of any neuron in the hidden layer of the information transient transfer model is determined based on the output of other neurons at the target moment, and the target moment is determined based on the external input and the inhibition strength of any neuron.

2. The method for transient transmission of brain-like information based on a neural network according to claim 1, characterized in that: The information transient transmission model includes: an input layer, a hidden layer and an output layer; Among them, the hidden layer includes: sensory brain areas, cognitive brain areas and decision-making brain areas. The sensory brain areas, cognitive brain areas and decision-making brain areas are hierarchically connected, and the hierarchical connection is used to simulate cross-brain area information transmission in the brain.

3. The method for transient transmission of brain-like information based on neural network according to claim 2, characterized in that: The neurons in the hidden layer of the information transient transmission model include: excitatory neurons and inhibitory neurons; Wherein, the excitatory neurons are used to simulate the excitatory synapses of the brain, and the inhibitory neurons are used to simulate the inhibitory synapses of the brain.

4. The method for transient transmission of brain-like information based on a neural network according to claim 1, characterized in that: The method further comprises: The firing rate of the target neuron at the current moment is determined according to the firing state of the target neuron at the previous moment, the firing states of other neurons at the previous moment, the external input received by the target neuron, and the inhibition strength of the target neuron at the current moment.

5. The method for transient transmission of brain-like information based on neural network according to claim 1, characterized in that: Before determining the firing rate of the target neuron at the current moment, the method further includes: The inhibition strength of the target neuron at the previous moment is determined according to the inhibition strength of the target neuron at the current moment and the firing state of the target neuron at the current moment.

6. The method for transient transmission of brain-like information based on a neural network according to claim 1, characterized in that: Before simulating the process of transient transmission of coded information in the brain based on the input data through the information transient transmission model to obtain a fixed transmission path of dynamic neuron firing, the method further includes: Acquire input data samples of different categories and test data samples corresponding to the input data samples; Inputting the input data sample and the test data sample into a preset information transient transfer model respectively, and obtaining a training matching result output by the preset information transient transfer model; Determine a cross entropy between the training matching result and a preset correct matching result, and use the cross entropy as a loss function; Back propagation is performed based on the loss function to update the preset information transient transfer model to obtain a trained information transient transfer model.

7. A brain-like information transient transmission device based on a neural network, characterized in that: include: An acquisition module, used to acquire input data of different categories and test data corresponding to the input data; A simulation module, used to simulate the process of transient transmission of coded information in the brain based on the input data through an information transient transmission model, and obtain a fixed transmission path of dynamic neuronal firing; An output module is used to input the test data into the information transient transfer model and obtain a matching result of the information transient transfer model output through the fixed transfer path after a preset target time, wherein the matching result is used to indicate the category matching between the test data and the input data; wherein the information transient transfer model is constructed based on a recurrent neural network, and the output of any neuron in the hidden layer of the information transient transfer model is determined based on the outputs of other neurons at a target time, and the target time is determined based on external input and the inhibition strength of any neuron.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for transient transmission of brain-like information based on a neural network as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for transient transmission of brain-like information based on a neural network as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for transient transmission of brain-like information based on a neural network as described in any one of claims 1 to 6 is implemented.

Images