CN113487008A - Environment sensing method based on caenorhabditis elegans pulse neural network model - Google Patents
Environment sensing method based on caenorhabditis elegans pulse neural network model Download PDFInfo
- Publication number
- CN113487008A CN113487008A CN202110541232.8A CN202110541232A CN113487008A CN 113487008 A CN113487008 A CN 113487008A CN 202110541232 A CN202110541232 A CN 202110541232A CN 113487008 A CN113487008 A CN 113487008A
- Authority
- CN
- China
- Prior art keywords
- caenorhabditis elegans
- neural network
- network model
- neuron
- initial
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 241000244203 Caenorhabditis elegans Species 0.000 title claims abstract description 162
- 238000003062 neural network model Methods 0.000 title claims abstract description 136
- 238000000034 method Methods 0.000 title claims abstract description 33
- 210000002569 neuron Anatomy 0.000 claims abstract description 194
- 210000000653 nervous system Anatomy 0.000 claims abstract description 49
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 43
- 230000006870 function Effects 0.000 claims abstract description 38
- 238000012421 spiking Methods 0.000 claims abstract description 38
- 230000002068 genetic effect Effects 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 15
- 238000012546 transfer Methods 0.000 claims abstract description 14
- 230000008447 perception Effects 0.000 claims abstract description 13
- 230000007613 environmental effect Effects 0.000 claims abstract description 9
- 239000012528 membrane Substances 0.000 claims description 25
- 210000002364 input neuron Anatomy 0.000 claims description 17
- 238000011156 evaluation Methods 0.000 claims description 16
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 108090000623 proteins and genes Proteins 0.000 claims description 10
- 230000035772 mutation Effects 0.000 claims description 8
- 230000001537 neural effect Effects 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 230000002964 excitative effect Effects 0.000 claims description 6
- 230000000284 resting effect Effects 0.000 claims description 6
- 230000002401 inhibitory effect Effects 0.000 claims description 5
- 229910001415 sodium ion Inorganic materials 0.000 claims description 5
- 229910001414 potassium ion Inorganic materials 0.000 claims description 4
- 108091026890 Coding region Proteins 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000013526 transfer learning Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 abstract description 15
- 230000002829 reductive effect Effects 0.000 abstract description 6
- 241000244206 Nematoda Species 0.000 description 28
- 230000006399 behavior Effects 0.000 description 18
- 210000004205 output neuron Anatomy 0.000 description 17
- 210000002161 motor neuron Anatomy 0.000 description 11
- 210000001044 sensory neuron Anatomy 0.000 description 8
- 238000005457 optimization Methods 0.000 description 7
- 230000000638 stimulation Effects 0.000 description 7
- 238000002474 experimental method Methods 0.000 description 5
- 230000033001 locomotion Effects 0.000 description 5
- 230000007246 mechanism Effects 0.000 description 4
- 210000003205 muscle Anatomy 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- 102000004257 Potassium Channel Human genes 0.000 description 2
- 108010052164 Sodium Channels Proteins 0.000 description 2
- 102000018674 Sodium Channels Human genes 0.000 description 2
- FKNQFGJONOIPTF-UHFFFAOYSA-N Sodium cation Chemical compound [Na+] FKNQFGJONOIPTF-UHFFFAOYSA-N 0.000 description 2
- 230000002238 attenuated effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000002457 bidirectional effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 108020001213 potassium channel Proteins 0.000 description 2
- 230000000946 synaptic effect Effects 0.000 description 2
- 206010001497 Agitation Diseases 0.000 description 1
- NPYPAHLBTDXSSS-UHFFFAOYSA-N Potassium ion Chemical compound [K+] NPYPAHLBTDXSSS-UHFFFAOYSA-N 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000004791 biological behavior Effects 0.000 description 1
- 238000013529 biological neural network Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 210000000349 chromosome Anatomy 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 210000001153 interneuron Anatomy 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000010534 mechanism of action Effects 0.000 description 1
- 244000005700 microbiome Species 0.000 description 1
- 230000008450 motivation Effects 0.000 description 1
- 210000000663 muscle cell Anatomy 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/008—Artificial life, i.e. computing arrangements simulating life based on physical entities controlled by simulated intelligence so as to replicate intelligent life forms, e.g. based on robots replicating pets or humans in their appearance or behaviour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Neurology (AREA)
- Robotics (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention provides an environment perception method based on a caenorhabditis elegans impulse neural network model, which comprises the steps of obtaining nervous system data, an input state sequence and a target sequence of the caenorhabditis elegans; simulating the connection relation of neurons in the caenorhabditis elegans nervous system, and constructing an initial caenorhabditis elegans impulse neural network model; and (3) using an optimized genetic algorithm, iteratively training an initial caenorhabditis elegans spiking neural network model by setting the cross probability and taking the target sequence as a learning target so as to enable the model to learn the characteristics of the input sequence until reaching the maximum evolution algebra, and obtaining the trained caenorhabditis elegans spiking neural network model, thereby determining a perception model of the biological robot for perceiving the environmental information. The pulse neuron model meets the characteristic that the biological neurons dynamically change along with time, has higher biological rationality, and has simpler transfer function and greatly reduced calculated amount.
Description
Technical Field
The invention belongs to the technical field of intelligence, and particularly relates to an environment perception method based on a caenorhabditis elegans impulse neural network model.
Background
The biological robot, a miniature robot, is often used to acquire natural environment information and enter into an environment where large, medium or small living beings cannot enter, for the purpose of sensing the surrounding environment.
The existing biological robot often uses a neural network model to learn the characteristics of microorganisms, and then simulates some behaviors of the organisms to realize disguised perception of the surrounding environment. Neural networks are considered to be the main motivation for the development of artificial intelligence, from the first perceptrons to the most popular deep neural networks today. The neuron model widely used in the deep neural network at present is an M-P neuron model, which is a multi-input single-output structure, and in the deep neural network, a feedforward connection network structure is generally used to simplify the complex information transfer in the biological neural network into a hierarchical transfer mode.
However, the deep neural network model still cannot accurately simulate the operation mechanism of the biological nervous system. The main reasons are two reasons: firstly, the biological nervous system is a very complex dynamic system, has the characteristics of complex topological structure, high nonlinearity, dynamic change and the like, and the detailed mechanism of the operation of the biological nervous system is not fully known and understood at present; secondly, the state of the actual biological neurons can change along with time, and the biological neurons are not the characteristics of hierarchical transmission of modern neural networks and do not completely simulate biological nervous systems. Therefore, the biological robot established by the neural network model used in the prior art cannot accurately simulate the biological behavior and is not beneficial to sensing environmental information.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides an environment sensing method based on a caenorhabditis elegans impulse neural network model. The technical problem to be solved by the invention is realized by the following technical scheme:
the invention provides an environment perception method based on a caenorhabditis elegans impulse neural network model, which comprises the following steps:
step 1: acquiring nervous system data, an input state sequence and a target sequence of caenorhabditis elegans;
the neural system data comprises the number of neurons, the types of the neurons and the connection relation among the neurons;
step 2: simulating the connection relation of neurons in a caenorhabditis elegans nervous system, and constructing a caenorhabditis elegans impulse neural network model framework and initial parameters of the framework to obtain an initial caenorhabditis elegans impulse neural network model;
the number of neurons in the caenorhabditis elegans spiking neural network architecture is the same as that of neurons in neuron data, a parameter exists among the neurons in the caenorhabditis elegans spiking neural network model architecture, and the connection mode of the neurons is the same as that of the neurons in the neuron data of the caenorhabditis elegans;
and step 3: inputting the input state sequence into an initial caenorhabditis elegans pulse neural network model, using an optimized genetic algorithm, reducing the cross probability of the individual number in the genetic algorithm by setting the cross probability attenuation which meets the requirement of an energy parameter search space at the initial training stage and at the later training stage, and iteratively training the initial caenorhabditis elegans pulse neural network model by taking the target sequence as a learning target so that the initial caenorhabditis elegans pulse neural network model learns the characteristics of the input state sequence until the maximum evolution algebra is reached to obtain the trained caenorhabditis elegans pulse neural network model;
and 4, step 4: and determining the trained caenorhabditis elegans impulse neural network model as a perception model of the biological robot for perceiving environmental information.
Optionally, step 2 includes:
step 21: setting a first neuron of a caenorhabditis elegans pulse neural network model so that the first neuron corresponds to a second neuron in a neural system one to one;
step 22: for each second neuron, determining the connection mode of the second neuron and other second neurons;
step 23: aiming at a first neuron corresponding to a second neuron, connecting the first neuron with other first neurons in the same way as the second neuron with other second neurons to obtain an initial caenorhabditis elegans pulse neural network model architecture;
step 24: simulating parameters in a nervous system, and setting initial parameters of an initial caenorhabditis elegans pulse neural network model architecture.
Optionally, step 3 includes:
step 31: inputting the input state sequence into an initial caenorhabditis elegans spiking neural network model, and taking a target sequence as a learning target of the initial caenorhabditis elegans spiking neural network model to obtain an output sequence;
step 32: carrying out gene coding on each parameter in the initial caenorhabditis elegans pulse neural network model to obtain a gene coding sequence of each parameter;
step 33: taking a difference function of an output sequence of the initial caenorhabditis elegans pulse neural network model and a target sequence as an evaluation function;
step 34: setting heritage parameters of a genetic algorithm;
wherein the genetic parameters include: the method comprises the following steps of (1) obtaining an initial population number L, a maximum evolution algebra G, a variation probability sigma, an initial cross probability gamma and a cross probability attenuation coefficient delta;
step 35: randomly generating L individuals as an initial population P0；
Wherein each individual represents all gene coding sequences in the initial caenorhabditis elegans impulse neural network model;
step 36: for initial population P0Every two individuals are crossed by a cross probability gamma, and filial generations obtained after crossing are added into an initial population P0When individual mutations are present, the mutated individuals are added to the starting population P0Obtaining a population after cross variation;
step 37: sorting the individuals in the population after the cross variation from big to small according to the fitness, and selecting the first L individuals to form a new population;
step 38: repeating the steps 36 to 37 until the evolution algebra of the population reaches the maximum evolution algebra;
step 39: and selecting parameters corresponding to individuals in the population reaching the maximum evolution algebra according to the fitness as parameters of the initial caenorhabditis elegans pulse neural network model to obtain the trained caenorhabditis elegans pulse neural network model.
Optionally, step 31 includes:
step 311: constructing a transfer function of a neuron in an initial caenorhabditis elegans pulse neural network model;
step 312: inputting the input state sequence into the initial caenorhabditis elegans spiking neural network model, and taking the target sequence as a learning target of the initial caenorhabditis elegans spiking neural network model;
step 313: determining activated neurons and inactivated neurons of the initial caenorhabditis elegans pulse neural network model in the learning process according to the transfer function;
step 314: and obtaining an output sequence obtained after the activated neuron transfer learning characteristics of the initial caenorhabditis elegans pulse neural network model.
Optionally, the transfer function is:
wherein u iscRepresents the membrane voltage of the neuron c,
discharge time constant, ENaRepresents the equilibrium potential of sodium ions, EKRespectively represent the equilibrium potential of potassium ions, -E is the resting voltage of the neuronal membrane voltage, xiRepresenting the output of neuron i, xjRepresents the output of neuron j, ωicRepresenting the connection coefficient, ω, of neuron i to neuron cjcThe absolute value of (A) represents the connection strength of neuron i to neuron c, PcRepresenting the collection of excitatory input neurons, Q, received by neuron ccRepresenting a collection of inhibitory input neurons.
Optionally, step 37 includes:
step 371: evaluating the individual fitness in the population after the cross variation by using an evaluation function;
step 372: and (4) sequencing the individuals in the population after the cross variation from large to small according to the fitness, and selecting the first L individuals to form a new population.
Optionally, step 39 includes:
step 391: evaluating the individual fitness in the population reaching the maximum evolution algebra by using an evaluation function;
step 392: and selecting the parameters corresponding to the individuals with the highest fitness as the parameters of the initial caenorhabditis elegans impulse neural network model to obtain the trained caenorhabditis elegans impulse neural network model.
Wherein, the number of the neurons is 302.
The invention provides an environment perception method based on a caenorhabditis elegans impulse neural network model, which comprises the steps of obtaining nervous system data, an input state sequence and a target sequence of the caenorhabditis elegans; simulating the connection relation of neurons in a caenorhabditis elegans nervous system, and constructing an initial caenorhabditis elegans impulse neural network model; and (3) using an optimized genetic algorithm, iteratively training an initial caenorhabditis elegans spiking neural network model by setting the cross probability and taking the target sequence as a learning target so as to enable the model to learn the characteristics of the input sequence until reaching the maximum evolution algebra, and obtaining the trained caenorhabditis elegans spiking neural network model, thereby determining a perception model of the biological robot for perceiving the environmental information. The pulse neuron model meets the characteristic that the biological neurons dynamically change along with time, has higher biological rationality, and has simpler transfer function and greatly reduced calculated amount.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
FIG. 1 is a schematic flow chart of an environment sensing method based on a caenorhabditis elegans spiking neural network model according to an embodiment of the present invention;
FIG. 2 is a diagram showing the connection relationship between neurons in a caenorhabditis elegans spiking neural network model according to an embodiment of the present invention;
FIG. 3a is a graph of the change in membrane voltage provided by an embodiment of the present invention;
FIG. 3b is a diagram illustrating the variation of the output pulse square wave generated by the membrane voltage according to the embodiment of the present invention;
FIG. 4 is a flow chart of calculating a neuron membrane voltage provided by an embodiment of the invention;
FIG. 5 is a mean evaluation function F of the population over 1000 iterationsavgThe convergence graph of (a).
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
Example one
As shown in FIG. 1, the method for sensing the environment based on the caenorhabditis elegans spiking neural network model provided by the invention comprises the following steps:
step 1: acquiring nervous system data, an input state sequence and a target sequence of caenorhabditis elegans;
the neural system data comprises the number of neurons, the types of the neurons and the connection relation among the neurons;
the cook team in 2019 determined the connection relationships and connection patterns between all neurons in the caenorhabditis elegans nervous system. According to the research result of the Steven j. cook team, there are 302 neurons in the nematode nervous system, and the connection mode between the neurons is only chemical connection and gap connection. The information transmission of the chemical connection has unidirectionality, so the unidirectional connection is used for simulating the chemical connection, and the bidirectional connection is used for simulating the gap connection with information bidirectional connectivity. For the two-way connection of the caenorhabditis elegans, the connection strength in two directions of the caenorhabditis elegans is assumed to be mutually independent, the two-way connection is divided into two mutually irrelevant one-way connections, and then the pulse neural network is composed of 302 neurons and 5907 one-way connections. The key to modeling the biological nervous system is to ensure that the model conforms to the characteristics of the biological nervous system, so the model constructed for the caenorhabditis elegans nervous system should satisfy the following three characteristics: firstly, because biological neurons dynamically change, a dynamic neuron model is adopted to model the neurons of the nematodes; secondly, the connection relation between the neurons should use the connection relation between the neurons in a real nematode nervous system, and the nematode nervous system has a very complex structure, so the established model is a feedback type neural network; and thirdly, the complete nematode nervous system should be modeled due to the wide cross connectivity of the nematode nervous system.
Step 2: simulating the connection relation of neurons in a caenorhabditis elegans nervous system, and constructing a caenorhabditis elegans impulse neural network model framework and initial parameters of the framework to obtain an initial caenorhabditis elegans impulse neural network model;
the number of neurons in the caenorhabditis elegans spiking neural network architecture is the same as that of neurons in neuron data, a parameter exists among the neurons in the caenorhabditis elegans spiking neural network model architecture, and the connection mode of the neurons is the same as that of the neurons in the neuron data of the caenorhabditis elegans; the number of neurons was 302.
As an optional embodiment of the present invention, step 2 includes:
step 21: setting a first neuron of a caenorhabditis elegans pulse neural network model so that the first neuron corresponds to a second neuron in a neural system one to one;
step 22: for each second neuron, determining the connection mode of the second neuron and other second neurons;
step 23: aiming at a first neuron corresponding to a second neuron, connecting the first neuron with other first neurons in the same way as the second neuron with other second neurons to obtain an initial caenorhabditis elegans pulse neural network model architecture;
step 24: simulating parameters in a nervous system, and setting initial parameters of an initial caenorhabditis elegans pulse neural network model architecture.
It is understood that in the nervous system of nematode, neurons are classified into three types, sensory neurons, interneurons, and motor neurons according to their functions. The sensory neurons are responsible for converting various types of stimuli of the external environment into electrical signals which are fed into the nervous system, and the electrical signals are equivalent to the input of the nervous system of nematodes. The nervous system of the nematode controls the muscle of the nematode through the motor neuron to enable the nematode to be in a corresponding motion state, so that the motion state of the nematode can be converted into an output state of the motor neuron, and the motor neuron is equivalent to the output of the nervous system of the nematode. The caenorhabditis elegans impulse neural network model designed by the invention comprises an input neuron and an output neuron which respectively correspond to a sensory neuron which is responsible for sensing certain stimulation in a caenorhabditis elegans nervous system and a motor neuron which controls the caenorhabditis elegans to generate behaviors corresponding to the input stimulation.
The caenorhabditis elegans pulse neural network model architecture is established, the model comprises 5907 parameters, the size of the parameters is limited between-1 and 1, the absolute value of the parameters represents the size of the connection strength between neurons, if the parameters are larger than 0, the connection belongs to excitatory connection, and if the parameters are smaller than 0, the connection belongs to inhibitory connection. The connection relationship between neurons in the model is shown in fig. 2, the abscissa is the neuron number for sending information, and the ordinate is the neuron number for receiving information. If there is a point at the location of coordinates (a, b) in fig. 2, there is a connection from neuron number a to neuron number b.
Table 1 is a numbering table for nematode neurons, 302 neurons of the nematode being numbered from 1 to 302. FIG. 2 shows the following four features of the nematode nervous system: firstly, the pharyngeal nervous system of the nematode is relatively independent, the number of the neuron which is numbered as the first 20 is the pharyngeal neuron of the nematode, and the pharyngeal neuron basically does not carry out signal transmission with other neurons; secondly, most neurons sending information are sensory neurons and intermediate neurons, and the distribution reflected as the left half points in fig. 2 is denser; thirdly, the information exchange among the motor neurons is more, the neurons with the numbers of 222 to 292 are basically the motor neurons for controlling the body wall muscles of the nematodes, and the information exchange among the neurons can enable the movement of the nematodes to be more coordinated; the points in the graph are distributed substantially symmetrically on both sides of the diagonal, because the bi-directional connection is decomposed into two unidirectional connections during the modeling process, and the two unidirectional connections appear as two points symmetrical about the diagonal on the graph.
The caenorhabditis elegans pulse neural network model architecture established by the invention has the following characteristics:
(1) the state of the neuron in the neural network model is changed along with time, when the membrane voltage of the neuron does not reach a threshold value, the neuron does not output a signal, and when the membrane voltage reaches the threshold value, the neuron outputs a pulse square wave signal.
(2) The neural network model completely simulates the connection relation between the neurons in the caenorhabditis elegans nervous system. The caenorhabditis elegans neural network model has the characteristics of having a large number of ring structures and feedback connection and conforming to the structure of the caenorhabditis elegans nervous system.
(3) The neural network model described above contains all 302 neurons of caenorhabditis elegans and all synaptic connections between neurons, and the model has a structural cross-connectivity of the caenorhabditis elegans nervous system.
TABLE 1 caenorhabditis elegans neuron numbering
And step 3: inputting an input state sequence into an initial caenorhabditis elegans pulse neural network model, using an optimized genetic algorithm, reducing the cross probability of the number of individuals in the genetic algorithm by setting the cross probability attenuation which meets the requirement of an energy parameter search space at the initial training stage and at the later training stage, wherein the target sequence is a learning target, and iteratively training the initial caenorhabditis elegans pulse neural network model to enable the initial caenorhabditis elegans pulse neural network model to learn the characteristics of the input state input sequence until the maximum evolution algebra is reached to obtain the trained caenorhabditis elegans pulse neural network model;
the caenorhabditis elegans spiking neural network model operates in a manner that how the output of the model is calculated given the inputs to the model. Because the connection mode of the neural network model is different from the common hierarchical feedforward connection, the operation mode of the network is also different from the common neural network, and the operation mode is defined as:
let the input neuron set of the model be A, the output neuron set be B, all the nervesThe set of elements is N, the model run time length is T, and the input and output of the network are represented by a sequence of length T, respectively. Wherein the input sequence consists of a set of input neuron states at each time t and the output sequence consists of a set of output neuron states at each time t. The state of the neuron in the model at the time t is divided into two states, namely a resting state and an activated state. If the input neuron a is stimulated by the outside at the moment t, the neuron is in an activated state, namely ξa(t) ═ 1, otherwise the neuron is in a resting state, ξa(t) is 0. If the output neuron b outputs the pulse square wave at the time t, the output neuron is in an activated state, namely ξb(t) ═ 1, otherwise the neuron is in a resting state, ξb(t) is 0. S (t) represents the set of all input neuron states of the model at time t, s (t) { ξ +a(t) | a ∈ a }, m (t) represents the set of all output neuron states of the model at time t, m (t) ═ ξb(t) | B ∈ B }. Then the input state sequence is { s (T) }, T0, 1, 2.., T-1 and the output state sequence is { m (T) }, T0, 1, 2.., T-1. The input state sequence is referred to as input sequence for short, and the output state sequence is referred to as output sequence for short.
And 4, step 4: and determining the trained caenorhabditis elegans impulse neural network model as a perception model of the biological robot for perceiving environmental information.
The invention provides an environment perception method based on a caenorhabditis elegans impulse neural network model, which comprises the steps of obtaining nervous system data, an input state sequence and a target sequence of the caenorhabditis elegans; simulating the connection relation of neurons in a caenorhabditis elegans nervous system, and constructing an initial caenorhabditis elegans impulse neural network model; inputting an input state sequence into an initial caenorhabditis elegans pulse neural network model, using an optimized genetic algorithm, reducing the cross probability of the number of individuals in the genetic algorithm by setting the cross probability attenuation which meets the requirement of an energy parameter search space at the initial training stage and at the later training stage, taking a target sequence as a learning target, and iteratively training the initial caenorhabditis elegans pulse neural network model to enable the initial caenorhabditis elegans pulse neural network model to learn the characteristics of the input state input sequence until the maximum evolution algebra is reached, so as to obtain the trained caenorhabditis elegans pulse neural network model, thereby determining a perception model of the biological robot for perceiving environmental information. The pulse neuron model meets the characteristic that the biological neurons dynamically change along with time, has higher biological rationality, and has simpler transfer function and greatly reduced calculated amount.
As an optional embodiment of the present invention, step 3 includes:
step 31: inputting the input state sequence into an initial caenorhabditis elegans spiking neural network model, and taking a target sequence as a learning target of the initial caenorhabditis elegans spiking neural network model to obtain an output sequence;
it is understood that although many results have been achieved in the study of the nervous system of caenorhabditis elegans, it is still impossible to understand how the nervous system of caenorhabditis elegans controls the muscle of the nematode to produce various behaviors, since the strength of the connections and the excitability of synaptic connections between neurons in the nervous system are not known. In the nervous system of the caenorhabditis elegans, sensory neurons are responsible for sensing external environment stimulation and then transmitting the stimulation into the nervous system, and corresponding motor neuron output is formed after treatment, so that various muscle cells and organs of the caenorhabditis elegans are controlled to generate various behaviors. According to the invention, environmental stimulation of the caenorhabditis elegans is simulated as an input sequence of a caenorhabditis elegans pulse neural network model, movement of the caenorhabditis elegans generated by the environmental stimulation is simulated as an output sequence, and behavior of the caenorhabditis elegans can be characterized by the two sequences.
In order to make the caenorhabditis elegans impulse neural network model have the ability of the nematode nervous system to control the muscle generation behavior, the behavior of the nematode is first modeled. Input neurons and output neurons of the spiking neural network model are determined from sensory neurons and motor neurons involved in nematode behavior. And determining an input sequence and a target output sequence according to the state change rule of the sensory neuron and the motor neuron in the movement process of the nematode. From the perspective of computational science, the behavior learning task of the neural network model is as follows: parameters in the caenorhabditis elegans impulse neural network model are optimized given the input sequence and the target output sequence of the model, so that the output sequence generated by the network is continuously fitted to the target output sequence.
Step 32: carrying out gene coding on each parameter in the initial caenorhabditis elegans pulse neural network model to obtain a gene coding sequence of each parameter;
step 33: taking a difference function of an output sequence of the initial caenorhabditis elegans pulse neural network model and a target sequence as an evaluation function;
step 34: setting heritage parameters of a genetic algorithm;
wherein the genetic parameters include: the method comprises the following steps of (1) obtaining an initial population number L, a maximum evolution algebra G, a variation probability sigma, an initial cross probability gamma and a cross probability attenuation coefficient delta;
step 35: randomly generating L individuals as an initial population P0；
Wherein each individual represents all gene coding sequences in the initial caenorhabditis elegans impulse neural network model;
step 36: for initial population P0Every two individuals are crossed by a cross probability gamma, and filial generations obtained after crossing are added into an initial population P0When individual mutations are present, the mutated individuals are added to the starting population P0Obtaining a population after cross variation;
step 37: sorting the individuals in the population after the cross variation from big to small according to the fitness, and selecting the first L individuals to form a new population;
step 38: repeating the steps 36 to 37 until the evolution algebra of the population reaches the maximum evolution algebra;
step 39: and selecting parameters corresponding to individuals in the population reaching the maximum evolution algebra according to the fitness as parameters of the initial caenorhabditis elegans pulse neural network model to obtain the trained caenorhabditis elegans pulse neural network model.
For the caenorhabditis elegans spiking neural network model provided by the invention, 5907 parameters needing to be optimized need to be explored, and a larger search space needs to be explored to determine the optimal combination of the parameter values, so that the output sequence of the caenorhabditis elegans spiking neural network model is the same as the target output sequence. In genetic algorithms, the purpose of crossover and mutation is to explore a larger parameter space and avoid the algorithm from converging on local extreme points. A genetic algorithm that involves only the selection operation will eventually converge to the best adapted individual of the initial population without further change. Meanwhile, the diversity of the population is reduced to 0 when the genetic algorithm comprising a selection mechanism and cross operation among individuals is evolved to a certain generation, and the algorithm falls into a local extreme point. The genetic algorithm can produce good effects only if the selection mechanism, the inter-individual cross and the individual variation operation are simultaneously provided. However, when the genetic algorithm is applied to optimize the parameters of the neural network model, the offspring generated by the individual intersection is generally low in fitness when the fitness of individuals in a population is generally high in the later stage of the algorithm, so that cross operation is not used in the parameter optimization work of some caenorhabditis elegans neural network models, and a good parameter optimization effect is achieved. The caenorhabditis elegans spiking neural network model parameter optimization algorithm is provided based on a genetic algorithm, the caenorhabditis elegans spiking neural network model established by the method has 5907 parameters and a larger search space, so that a larger cross probability is used to ensure a sufficiently large parameter search space in the initial stage of the algorithm, and the cross probability is continuously attenuated due to the fact that cross operation destroys the characteristics found in the evolution process of the genetic algorithm in the later stage of the algorithm. The learning algorithm comprises the following specific steps:
(1) the gene encodes. A real number coding method is adopted to code 5907 parameters of the caenorhabditis elegans pulse neural network model. Between-1 and 1, real number encoding results in 5907 sequences, propagation functions,
(2) and selecting an evaluation function (the difference between the target sequence and the model output after the propagation of the propagation function) and setting parameters of the genetic algorithm. The parameters mainly include an initial population number L, a maximum evolution algebra G, a variation probability σ, an initial cross probability γ, and a cross probability attenuation coefficient δ, and it should be noted that a larger value should be selected for the initial cross probability. g denotes an algebraic counter, PgRepresents the g generation population.
(3) Randomly generating L piecesIndividuals as the initial population P0Let g be 0 at the same time.
One individual is 5907 parameters, one parameter is a chromosome in the individual
(4) Performing cross operation, crossing every two individuals in the population with a cross probability gamma, and adding the crossed offspring to PgIn (1).
(5) Performing mutation operation, wherein the population individual has probability of generating mutation, and if the mutation is performed, adding the mutated individual to PgIn (1).
(6) A selection operation is performed. The population PgAll the individuals in the population are ranked from high fitness to low fitness, only the first L individuals are reserved, and a new population P is formedg+1。
(7) It is determined whether a termination condition is reached. If G is equal to G, the algorithm ends and the population PgThe population obtained by final optimization; otherwise, let g be g +1 and γ be γ × δ, and continue to execute step (4).
The learning algorithm has the following two advantages: firstly, a large enough parameter search space in the early stage can be ensured by mutation operation and a large cross probability in the initial stage, and the method is suitable for a caenorhabditis elegans impulse neural network model with a large number of parameters; secondly, in the later stage of the algorithm, because the fitness of the offspring obtained by the crossover operation is generally low, the crossover probability is continuously attenuated until the value is 0. Only variation operation is included, so that the number of population individuals needing selection operation is greatly reduced, the running of the algorithm can be accelerated, and the retained variation operation can be continuously optimized.
As an alternative embodiment of the present invention, step 31 includes:
step 311: constructing a transfer function of a neuron in an initial caenorhabditis elegans pulse neural network model;
the purpose of designing neuron models is to simulate biological neurons, and therefore should satisfy the characteristics of biological neurons. The traditional M-P neuron model belongs to a static neuron model, the state of neurons does not change along with time, and the biological rationality is low. Compared with an M-P neuron model, the pulse neuron model has biological rationality, has the characteristics of biological neuron information integration and nonlinear processing, and only when the voltage of a neuron membrane reaches a threshold value, the neuron can generate an output signal to meet the characteristic of dynamic change of the state of the biological neuron. The H-H model is a classical pulse neuron model, describes the change of neuron membrane voltage and the generation mechanism of action potential through four differential equations, and has high biological rationality. However, the model is complex, which results in high computational complexity, and is not beneficial to subsequent simulation work. In the H-H model, two factors mainly influencing the neuron membrane voltage are the conductance of a sodium ion channel and the conductance of a potassium ion channel, wherein the increase of the sodium ion conductance increases the neuron membrane voltage, and the increase of the potassium ion conductance decreases the neuron membrane voltage.
The invention provides a pulse neuron model by compromising biological rationality and computational high efficiency, and simplifies nonlinear changes of conductance of sodium ion and potassium ion channels into linear changes which are respectively controlled by excitatory input and inhibitory input.
The transfer function of the model is:
wherein,
discharge time constant, ENa、EKRepresents the equilibrium potential of sodium ions and potassium ions, respectively, -E is the resting voltage of the neuronal membrane voltage. The reference value of the parameter is that,
ENa＝55mv，EK＝75mv，E＝69mv。ucrepresents the membrane voltage of neuron c once ucIs greater than a threshold voltage VthWhen the voltage of neuron membrane is restored to equilibrium state u at falling edge of square wave, neuron outputs a pulse square wave with voltage of 30mv lasting 4 unit timec-E. That is, the neuron is only at membrane voltageAn output is generated when the threshold is exceeded. x is the number ofiRepresenting the output of neuron i, xjRepresenting the output of neuron j. OmegaicRepresenting the connection coefficient, ω, of neuron i to neuron cicThe absolute value of (a) represents the connection strength of neuron i to neuron c, the sign thereof represents the excitement inhibition property, and the range of the connection coefficient ω is limited to-1 ≦ ω ≦ 1. Set PcAnd QcRepresenting the excitatory input neuron set and the inhibitory input neuron set, respectively, received by neuron c. If the connection coefficient omega of neuron i to neuron cicIf > 0, the linkage is excitatory, i.e., PcIf ω isicIf < 0, the connection is a suppressive connection, i.e., i ∈ Qc. Assume that the set of all input neurons of neuron c is NcThen satisfy Pc∪Qc＝NcAnd
fig. 3a is a schematic diagram of changes in membrane voltage and output voltage of the pulse neuron model. FIG. 3a is a graph of membrane voltage change, which abruptly increases to 30mv and outputs a 30mv pulsed square wave when the membrane voltage reaches a threshold, as shown in FIG. 3 b.
The flow of calculating the membrane voltage u (t) of each neuron at the time t in the nematode pulse neural network model is shown in figure 4. When the membrane voltage of each output neuron at the time T is obtained through calculation, the state set M (T) of all the output neurons at the time T is obtained at the same time, and after T unit times, the output sequence { M (T) }, T ═ 0,1, 2. As can be seen from fig. 4, the operation of the nematode spiking neural network model is characterized by updating the membrane voltages of all neurons at the same time, and entering the next time after the update is completed.
Step 312: inputting the input state sequence into the initial caenorhabditis elegans spiking neural network model, and taking the target sequence as a learning target of the initial caenorhabditis elegans spiking neural network model;
step 313: determining activated neurons and inactivated neurons of the initial caenorhabditis elegans pulse neural network model in the learning process according to the transfer function;
step 314: and obtaining an output sequence obtained after the activated neuron transfer learning characteristics of the initial caenorhabditis elegans pulse neural network model.
As an alternative embodiment of the present invention, step 37 includes:
step 371: evaluating the individual fitness in the population after the cross variation by using an evaluation function;
step 372: and (4) sequencing the individuals in the population after the cross variation from large to small according to the fitness, and selecting the first L individuals to form a new population.
As an alternative embodiment of the present invention, step 39 includes:
step 391: evaluating the individual fitness in the population reaching the maximum evolution algebra by using an evaluation function;
step 392: and selecting the parameters corresponding to the individuals with the highest fitness as the parameters of the initial caenorhabditis elegans impulse neural network model to obtain the trained caenorhabditis elegans impulse neural network model.
The performance of the sensing method provided by the invention is verified through simulation experiments.
The genetic algorithm is used for carrying out parameter optimization on the caenorhabditis elegans pulse neural network model, and the feasibility of carrying out parameter optimization on the model by the learning algorithm provided by the invention is demonstrated through experiments. The input sequence and the target output sequence of the caenorhabditis elegans pulse neural network model are generated by another parameter randomly generated caenorhabditis elegans pulse network model, and the randomly generated behavior is a potential behavior of the caenorhabditis elegans. The specific production steps are as follows:
(1) and generating a caenorhabditis elegans pulse neural network model with random parameters.
(2) Repeating the steps for 3 and 4 twice.
(3) Several sensory neurons were randomly selected as input neurons to the network and all 108 motor neurons were selected as output neurons to the network. Let the set of input neurons be a and the set of output neurons be B.
An input sequence of time length T is randomly generated (s (T)), where T is 0, 1.
(4) The state of the output neuron at each moment in time T is recorded, and a target output sequence { d (T) } with the length T being 0, 1.
Taking T-5, the input and output sequences of the two groups of caenorhabditis elegans pulse neural network models are generated in the manner. Wherein each set of input-output sequences can be regarded as a certain behavior of caenorhabditis elegans. In the experiment, the pulse neural network model is trained through the learning algorithm provided by the invention, so that the output sequence generated by the pulse neural network model under the stimulation of the input sequence is the same as the target output sequence, which is equivalent to training the caenorhabditis elegans pulse neural network model to learn two behaviors at the same time. Since the nervous system of caenorhabditis elegans produces non-unitary behavior, this experiment also performed multi-behavior learning on this model. For the first randomly generated behavior, let the set of input neurons be A(1)The set of output neurons is B(1)The input sequence and the target output sequence are { S }(1)(t) } and { D(1)(T) }, wherein T ═ 0, 1. Similarly, for the second randomly generated behavior, let the set of input neurons be A(2)The set of output neurons is B(2)The input sequence and the target output sequence are { S }(2)(t) } and { D(2)(T) }, wherein T ═ 0, 1. The evaluation function in the learning algorithm is determined by the difference between the target output sequence { D (t)) } and the actual output sequence { M (t)) } of the model, wherein the smaller the difference is, the higher the fitness of the individual is, and the larger the difference is, the lower the fitness of the individual is. The invention defines a function F as an evaluation function of a genetic algorithm for measuring the difference between an actual output sequence and a target output sequence. For the first behavior, the target state of the output neuron i at time t is
The actual state is
For the second behavior, the target state of output neuron j at time tIs composed of
The actual state is
The expression of the evaluation function F is as follows:
wherein omega is the connection coefficient of the caenorhabditis elegans impulse neural network model and is a parameter to be learned. The larger the value of the function F is, the larger the difference between the target output sequence and the actual output sequence is, and the lower the fitness of the individual is represented; conversely, a lower value of the function F indicates a higher fitness of the individual.
The parameters of the learning algorithm are set as follows: the population number L is 50, the initial crossover probability γ is 0.5, the variation probability σ is 0.2, the maximum evolution generation G is 1000, and the crossover probability attenuation coefficient δ is 0.99. Average evaluation function value F of population used in this experimentavgTo measure the average fitness of the population. FavgA higher value of (A) indicates a lower fitness of the population, FavgA lower value of (a) indicates a higher fitness of the population.
FIG. 5 shows the mean evaluation function F of the population over 1000 iterationsavgThe convergence graph of (a). Since the learning algorithm only retains the optimal individuals during the selection process, the curve decreases monotonically. Objective function F of populationavgThe continuous reduction shows that the actual output sequence of the neural network model is closer to the target output sequence, and finally, only the output sequences of a few neurons in all the output neurons are different from the target output sequence, so that the learning algorithm provided by the invention has feasibility for learning parameters of the caenorhabditis elegans spiking neural network model.
The pulse neuron model provided by the invention meets the characteristic that biological neurons dynamically change along with time, and has higher biological rationality. Compared with an H-H pulse neuron model, the change process of the membrane voltage is described by only one differential equation, so that the calculated amount is greatly reduced, and the simulation is easier. The complete nervous system of the caenorhabditis elegans is modeled based on the pulse neuron model, and the connection mode among the neurons in the model is consistent with the real nervous system of the caenorhabditis elegans. Finally, aiming at the parameter optimization problem of the model, a learning algorithm based on a genetic algorithm is provided, and the feasibility of the learning algorithm for optimizing the parameters of the nematode impulse neural network model is verified through experiments.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.
Claims (8)
1. An environment perception method based on a caenorhabditis elegans pulse neural network model is characterized by comprising the following steps:
step 1: acquiring nervous system data, an input state sequence and a target sequence of caenorhabditis elegans;
the nervous system data comprises the number of neurons, the types of the neurons and connection relations among the neurons;
step 2: simulating the connection relation of neurons in the caenorhabditis elegans nervous system, and constructing a caenorhabditis elegans impulse neural network model framework and initial parameters of the framework to obtain an initial caenorhabditis elegans impulse neural network model;
the number of neurons in the caenorhabditis elegans spiking neural network architecture is the same as that in the neuron data, a parameter exists among the neurons in the caenorhabditis elegans spiking neural network model architecture, and the connection mode of the neurons is the same as that of the neurons in the neuron data of the caenorhabditis elegans;
and step 3: inputting the input state sequence into an initial caenorhabditis elegans spiking neural network model, using an optimized genetic algorithm, reducing the cross probability of the number of individuals in the genetic algorithm by setting the cross probability attenuation which meets the requirement of the parameter search space at the initial training stage and at the later training stage, and iteratively training the initial caenorhabditis elegans spiking neural network model by taking a target sequence as a learning target so that the initial caenorhabditis elegans spiking neural network model learns the characteristics of the input state sequence until the maximum evolution algebra is reached to obtain the trained caenorhabditis elegans spiking neural network model;
and 4, step 4: and determining the trained caenorhabditis elegans impulse neural network model as a perception model of the biological robot for perceiving environmental information.
2. The context aware method of claim 1, wherein the step 2 comprises:
step 21: setting a first neuron of a caenorhabditis elegans pulse neural network model so that the first neuron corresponds to a second neuron in the neural system one to one;
step 22: for each second neuron, determining the connection mode of the second neuron and other second neurons;
step 23: aiming at a first neuron corresponding to a second neuron, connecting the first neuron with other first neurons in the same way as the second neuron with other second neurons to obtain an initial caenorhabditis elegans pulse neural network model architecture;
step 24: and simulating parameters in the nervous system, and setting initial parameters of the initial caenorhabditis elegans pulse neural network model architecture.
3. The context aware method of claim 1, wherein the step 3 comprises:
step 31: inputting the input state sequence into an initial caenorhabditis elegans spiking neural network model, and taking the target sequence as a learning target of the initial caenorhabditis elegans spiking neural network model to obtain an output sequence;
step 32: carrying out gene coding on each parameter in the initial caenorhabditis elegans pulse neural network model to obtain a gene coding sequence of each parameter;
step 33: taking a difference function of the output sequence of the initial caenorhabditis elegans pulse neural network model and the target sequence as an evaluation function;
step 34: setting heritage parameters of a genetic algorithm;
wherein the genetic parameters include: the method comprises the following steps of (1) obtaining an initial population number L, a maximum evolution algebra G, a variation probability sigma, an initial cross probability gamma and a cross probability attenuation coefficient delta;
step 35: randomly generating L individuals as an initial population P0；
Wherein each individual represents all gene coding sequences in the initial caenorhabditis elegans impulse neural network model;
step 36: for the initial population P0Every two individuals are crossed by a cross probability gamma, and filial generations obtained after crossing are added into the initial population P0When individual mutations are present, the mutated individuals are added to the starting population P0Obtaining a population after cross variation;
step 37: sorting the individuals in the population after the cross variation from large to small according to the fitness, and selecting the first L individuals to form a new population;
step 38: repeating the steps 36 to 37 until the evolution algebra of the population reaches the maximum evolution algebra;
step 39: and selecting parameters corresponding to individuals in the population reaching the maximum evolution algebra according to the fitness as parameters of the initial caenorhabditis elegans pulse neural network model to obtain the trained caenorhabditis elegans pulse neural network model.
4. The context aware method of claim 3, wherein the step 31 comprises:
step 311: constructing a transfer function of a neuron in the initial caenorhabditis elegans pulse neural network model;
step 312: inputting the input state sequence into an initial caenorhabditis elegans spiking neural network model, and taking the target sequence as a learning target of the initial caenorhabditis elegans spiking neural network model;
step 313: determining activated neurons and inactivated neurons of the initial caenorhabditis elegans spiking neural network model in a learning process according to the transfer function;
step 314: and obtaining an output sequence obtained after the activated neuron transfer learning characteristics of the initial caenorhabditis elegans pulse neural network model.
5. The context awareness method of claim 4, wherein the transfer function is:
wherein u iscRepresents the membrane voltage of the neuron c,
discharge time constant, ENaRepresents the equilibrium potential of sodium ions, EKRespectively represent the equilibrium potential of potassium ions, -E is the resting voltage of the neuronal membrane voltage, xiRepresenting the output of neuron i, xjRepresents the output of neuron j, ωicRepresenting the connection coefficient, ω, of neuron i to neuron cjcThe absolute value of (A) represents the connection strength of neuron i to neuron c, PcRepresenting the collection of excitatory input neurons, Q, received by neuron ccRepresenting a collection of inhibitory input neurons.
6. The context aware method of claim 4, wherein the step 37 comprises:
step 371: evaluating the individual fitness in the population after the cross variation by using the evaluation function;
step 372: and sequencing the individuals in the population after the cross variation from large to small according to the fitness, and selecting the first L individuals to form a new population.
7. The context aware method of claim 4, wherein the step 39 comprises:
step 391: evaluating the individual fitness in the population reaching the maximum evolution algebra by using the evaluation function;
step 392: and selecting the parameters corresponding to the individuals with the highest fitness as the parameters of the initial caenorhabditis elegans impulse neural network model to obtain the trained caenorhabditis elegans impulse neural network model.
8. The context awareness method of any one of claims 1-6, wherein the number of neurons is 302.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110541232.8A CN113487008A (en) | 2021-05-18 | 2021-05-18 | Environment sensing method based on caenorhabditis elegans pulse neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110541232.8A CN113487008A (en) | 2021-05-18 | 2021-05-18 | Environment sensing method based on caenorhabditis elegans pulse neural network model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113487008A true CN113487008A (en) | 2021-10-08 |
Family
ID=77932857
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110541232.8A Pending CN113487008A (en) | 2021-05-18 | 2021-05-18 | Environment sensing method based on caenorhabditis elegans pulse neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113487008A (en) |
-
2021
- 2021-05-18 CN CN202110541232.8A patent/CN113487008A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Noriega | Multilayer perceptron tutorial | |
| Jadav et al. | Optimizing weights of artificial neural networks using genetic algorithms | |
| US11514327B2 (en) | Apparatus and method for utilizing a parameter genome characterizing neural network connections as a building block to construct a neural network with feedforward and feedback paths | |
| CN116881683A (en) | GA-AM-GRU-based flow industrial energy consumption prediction method | |
| WO2019246116A1 (en) | Apparatus and method for utilizing a parameter genome characterizing neural network connections as a building block to construct a neural network with feedforward and feedback paths | |
| CN110853707A (en) | Gene regulation and control network reconstruction method based on deep learning | |
| Han et al. | A new approach for function approximation incorporating adaptive particle swarm optimization and a priori information | |
| CN113487008A (en) | Environment sensing method based on caenorhabditis elegans pulse neural network model | |
| Stromatias | Developing a supervised training algorithm for limited precision feed-forward spiking neural networks | |
| El-Dahshan et al. | Artificial neural network and genetic algorithm hybrid technique for nucleus–nucleus collisions | |
| Lin et al. | A hybrid evolutionary learning algorithm for TSK-type fuzzy model design | |
| Upadhyay et al. | ARTIFICIAL NEURAL NETWORKS: A REVIEW STUDY | |
| Wu et al. | Fuzzy modeling employing fuzzy polyploidy genetic algorithms | |
| Zhang et al. | Combat Decision-Making Modeling Method Based on Genetic Neural Network | |
| Mestre | Modeling C. elegans Nervous System’s Behavior using Machine Learning Techniques | |
| Hell et al. | Training Neurofuzzy Networks with Participatory Learning. | |
| Zhao et al. | Stochastic connection neural networks | |
| Hassan et al. | A spiking neural network controller design with VHDL based on SSO algorithm for inverted pendulum | |
| Chen | Optimization of neural network based on improved genetic algorithm | |
| Salama et al. | Evolving neural network controllers | |
| Okada | Interval-valued differential evolution for evolving neural networks with interval weights and biases | |
| Donaldson et al. | A system for evolving neural architectures | |
| Chauhan et al. | Soft Computing Methods | |
| Bouzerdoum et al. | A generalized feedforward neural network architecture and its training using two stochastic search methods | |
| Shen et al. | A method of image classification with optimized BP neural network by genetic algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20211008 |
|
| RJ01 | Rejection of invention patent application after publication |