CN109063319A - A kind of analogy method of bioecosystem neural network based - Google Patents
A kind of analogy method of bioecosystem neural network based Download PDFInfo
- Publication number
- CN109063319A CN109063319A CN201810843096.6A CN201810843096A CN109063319A CN 109063319 A CN109063319 A CN 109063319A CN 201810843096 A CN201810843096 A CN 201810843096A CN 109063319 A CN109063319 A CN 109063319A
- Authority
- CN
- China
- Prior art keywords
- block
- sub
- attention mechanism
- biology
- input
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/045—Combinations of networks
-
- 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
Abstract
A kind of analogy method of bioecosystem neural network based: establishing the data set of the analogy method of bioecosystem neural network based, including constitutes tag block, bamboo slips used for divination or drawing lots block 1, label sub-block 2, original block, initial sub-block 1 and initial sub-block 2；Initial sub-block 1 is input in sub-network 1, influence of the model ecosystem to biology；Original block is input in sub-network 2, the inter-species for simulating biology influences；Initial sub-block 2 is input in sub-network 3, simulating in the kind of biology influences；By similarity matrix 1, similarity matrix 2 and similarity matrix 3 are averaging by corresponding element, average similarity matrix are obtained, then all elements of average similarity matrix are averaging, to obtain similarity score.The data that the present invention inputs are the initial cell of biology, and the process of the training iteration of neural network is the process that biology gradually develops and trained neural network is the biological ecosystem.
Description
Technical field
The present invention relates to a kind of analogy methods of bioecosystem.More particularly to a kind of biology neural network based
The analogy method of the ecosystem.
Background technique
With computer hardware, internet, the progress of big data and neural network, machine learning and deep learning rise abruptly rapidly
It rises and develops.Machine learning and deep learning are in image classification, video frequency abstract, image recognition, speech recognition, image retrieval, word
Curtain generates and there is preferable effect in the directions such as personalized search.In these directions and field, occur much doing well
Neural network model, such as ImgNet, VggNet, GoogleNet, LSTM and CapsNet etc..
Neural network continuously adjusts the parameter of network internal, by constantly iterative learning to constantly optimize
Entire neural network.The inner parameter of neural network be exactly the network to the various changes linearly or nonlinearly of input data,
It influences and limits.
Summary of the invention
The technical problem to be solved by the invention is to provide one kind constantly to deposit close to real world during training
Biological situation bioecosystem neural network based analogy method.
The technical scheme adopted by the invention is that: a kind of analogy method of bioecosystem neural network based, packet
Include following steps:
1) data set of the analogy method of bioecosystem neural network based is established, comprising: every kind of biology of setting
Attribute, the coding of attribute vector and division simultaneously constitute tag block, bamboo slips used for divination or drawing lots block 1, label sub-block 2, original block, 1 and of initial sub-block
Initial sub-block 2；
2) initial sub-block 1 is input in sub-network 1, model ecosystem to biology influence, the sub-network 1 by
CapsNet is constituted；
3) original block is input in sub-network 2, the inter-species for simulating biology influences, and the sub-network 2 is by LSTM and mutually
Attention mechanism module is constituted；
4) initial sub-block 2 is input in sub-network 3, simulating in the kind of biology influences, the sub-network 3 be by LSTM and
It is constituted from attention mechanism module；
5) similarity matrix 1, similarity matrix 2 and similarity matrix 3 are averaging by corresponding element, are obtained averagely similar
Matrix is spent, then all elements of average similarity matrix are averaging, to obtain similarity score.
Every kind of biological attribute described in step 1), comprising: boundary a, door b, guiding principle c, mesh d, section e, belong to f, kind g, the ecosystem
Type h, it is given birth to across ecosystem quantity i, population quantity j, quality k, natural enemy biological species number l, natural enemy biomass m, predation
Species number n, biological total amount o is preyed on.
The coding of attribute vector described in step 1) and division simultaneously constitute tag block, bamboo slips used for divination or drawing lots block 1, label sub-block 2, initial
Block, initial sub-block 1 and initial sub-block 2, comprising:
(1) by every kind of biological combinations of attributes at an attribute vector A with 15 attributes1=a, b, c, d, e, f,
G, h, i, j, k, l, m, n, o }, circle a, door b, guiding principle c, mesh d, section e, category f, kind g are encoded step by step using the mode of binary tree；It will belong to
Property vector A1It is respectively divided into the attribute subvector A with 9 attributes2With the attribute subvector A with 6 attributes3, wherein A2
={ a, b, c, d, e, f, g, h, i } and A3={ j, k, l, m, n, o }；
(2) attribute vector of m × n kind biology is arranged in the tensor of m × n × 15, constitutes tag block；By m × n kind biology
Subvector A2And A3It is arranged together in the tensor of m × n × 9 and m × n × 6, respectively constitutes label sub-block 1 and label sub-block 2；
(3) tensor of m × n × 15 for being arranged in m × n identical monadic attribute vectors constitutes original block；
M × n identical the tensors of monadic m × n × 15 are divided into the tensor of m × n × 9 and m × n × 6, are respectively constituted just
Beginning sub-block 1 and initial sub-block 2.
Step 2) includes: that the output result and 1 corresponding position of label sub-block of CapsNet are calculated separately according to the type of biology
The cosine similarity of attribute vector obtains similarity matrix 1；The output result of CapsNet is cyclically input in CapsNet
It is trained and passes through the constantly parameter of readjustment CapsNet of loss function 1.
Step 3) includes:
(1) original block is input in LSTM, the output result of LSTM is input to mutual attention mechanism module, it is described
Mutual attention mechanism module includes mutual attention mechanism submodule 1 and mutual attention mechanism submodule 2, wherein the mutual note
Meaning power mechanism submodule 1 is made of m × n full articulamentums and Softmax layers, and the mutual attention mechanism submodule 2 is
By m × n, mutually attention weight modules are formed, the mutual attention weight module be by every kind of biological attribute vector m ×
N-1 weight vectors form, and the element inside each weight vectors is identical；
(2) in mutual attention mechanism submodule 1, by each biological species attribute vector be separately input to full articulamentum and
It is calculated in Softmax layers, then the calculated result of mutual attention mechanism submodule 1 is input to mutual attention mechanism submodule
In 2, every kind of biological attribute vector can distribute m × n-1 weight vectors, in mutual attention mechanism submodule 2, by each biology
Attribute vector respectively with distribute to this kind of m × n-1 biological weight vectors corresponding element and be multiplied, and m × n-1 calculating is tied
Fruit is multiplied with the attribute vector of other biological respectively, each biology obtain other all biological m × n-1 weighting attributes to
Amount, and weight attribute vector for m × n-1 and be added the attribute vector new as this kind of biology；
(3) the output result of mutual attention mechanism module and the cosine similarity of tag block are calculated, similarity matrix is obtained
2, the output result of LSTM is cyclically input in LSTM, is trained and by the constantly readjustment LSTM of loss function 2 and mutually
The parameter of attention mechanism module.
Step 4) includes:
(1) initial sub-block 2 is input in LSTM, the output result of LSTM is input to from attention mechanism module, certainly
Attention mechanism module includes from attention mechanism submodule 1 and from attention mechanism submodule 2；It is described from attention mechanism
Submodule 1 is made of the full articulamentums of m × n and Softmax layer, and described from attention mechanism submodule 2 is a by m × n
Weight vectors form, and the element inside each weight vectors is identical；
(2) from attention mechanism submodule 1, by each biological species attribute vector be separately input to full articulamentum and
It is calculated, then will be input to from the calculated result of attention mechanism submodule 1 from attention mechanism submodule in Softmax layers
In 2, from attention mechanism submodule 2, by each biological attribute vector and this kind of biological 1 weight vectors pair are distributed to
It answers element multiplication, and calculated result is multiplied the attribute vector new as this kind of biology with the attribute vector of the biology；
(3) cosine similarity for calculating the output result and label sub-block 2 from attention mechanism module, obtains similarity moment
Battle array 3, the output result of LSTM is cyclically input in LSTM and is trained, and by loss function 3 constantly readjustment LSTM and
From the parameter of attention mechanism module.
A kind of analogy method of bioecosystem neural network based of the invention, is parameter neural network based
Adjustment, circuit training optimize network and carry out various linear or heterogeneous linear influence and limitation to input data, to biology
Ecological succession and development are simulated.The data of input are the initial cell of biology, the training iteration of neural network
Process is the process that biology gradually develops and trained neural network is the biological ecosystem.The present invention has as follows
Feature:
(1) novelty: it has been put forward for the first time the simulation algorithm of bioecosystem neural network based, has utilized nerve net
The relevant knowledge of network, machine learning and deep learning probes into the evolution process of bioecosystem.
(2) validity: as the continuous iteration and parameter of neural network adjust, the unicellular life of the unification of initial input
Object can be eventually differentiated into the different biology in Different ecosystems existence, possess different quality, and colony number etc. is each
The biological attribute of kind various kinds, and constantly close to biology situation existing for real world during training.
(3) practicability: the evolution process and the ecosystem for being used to simulate biology for deep learning and neural network are gradually
The process of formation.By observing the adjustment variation of neural network parameter, it can analyze to obtain biology and environment, inter-species in training
With influencing each other in kind, the process with mutual game is mutually restricted, further appreciates that bioecosystem so as to closer.
Detailed description of the invention
Fig. 1 is the neural network structure schematic diagram of the analogy method of the bioecosystem the present invention is based on neural network；
Fig. 2 is the structural schematic diagram of mutual attention mechanism module；
Fig. 3 is the structural schematic diagram from attention mechanism module.
Specific embodiment
A kind of simulation of bioecosystem neural network based of the invention is calculated below with reference to embodiment and attached drawing
Method is described in detail.
The simulation algorithm of a kind of bioecosystem neural network based of the invention, it is intended to go to recognize using neural network
Know, the development and formation of analysis and simulation bioecosystem.
As shown in Figure 1, a kind of simulation algorithm of bioecosystem neural network based of the invention, including walk as follows
It is rapid:
1) data set of the simulation algorithm of bioecosystem neural network based is established, comprising: every kind of biology of setting
Attribute, the coding of attribute vector and division simultaneously constitute tag block, bamboo slips used for divination or drawing lots block 1, label sub-block 2, original block, 1 and of initial sub-block
Initial sub-block 2；
The biological attribute of described every kind, comprising: boundary a, door b, guiding principle c, mesh d, section e, belong to f, kind g, ecosystem type h,
Across ecosystem quantity i, population quantity j, quality k, natural enemy biological species number l, natural enemy biomass m, predation biological species
Number n, biological total amount o is preyed on；Wherein, boundary a, door b, guiding principle c, mesh d, section e, category f, kind g are the grade classification and name to biology；It is raw
State system type h is a main ecosystem existing for the biology, is that the biology can survive across ecosystem quantity i
The ecosystem total quantity；Population quantity j, quality k, natural enemy biomass m, predation biology total amount o are accurate to the order of magnitude
?；Natural enemy biological species number l, predation biological species number n are the species number of the biology of common natural enemy and predation.
The coding of the attribute vector and division simultaneously constitute tag block, bamboo slips used for divination or drawing lots block 1, label sub-block 2, original block, initial
Sub-block 1 and initial sub-block 2, comprising:
(1) by every kind of biological combinations of attributes at an attribute vector A with 15 attributes1=a, b, c, d, e, f,
G, h, i, j, k, l, m, n, o }, circle a, door b, guiding principle c, mesh d, section e, category f, kind g are encoded step by step using the mode of binary tree；It will belong to
Property vector A1It is respectively divided into the attribute subvector A with 9 attributes2With the attribute subvector A with 6 attributes3, wherein A2
={ a, b, c, d, e, f, g, h, i } and A3={ j, k, l, m, n, o }；
(2) attribute vector of m × n kind biology is arranged in the tensor of m × n × 15, constitutes tag block；By m × n kind biology
Subvector A2And A3It is arranged together in the tensor of m × n × 9 and m × n × 6, respectively constitutes label sub-block 1 and label sub-block 2；Tool
Body includes:
The output result and 1 corresponding position attribute vector of label sub-block of CapsNet are calculated separately according to the type of biology
Cosine similarity obtains similarity matrix 1；The output result of CapsNet is cyclically input in CapsNet and is trained simultaneously
Pass through the constantly parameter of readjustment CapsNet of loss function 1.
(3) tensor of m × n × 15 for being arranged in m × n identical monadic attribute vectors constitutes original block；
M × n identical the tensors of monadic m × n × 15 are divided into the tensor of m × n × 9 and m × n × 6, are respectively constituted just
Beginning sub-block 1 and initial sub-block 2.
2) initial sub-block 1 is input in sub-network 1, model ecosystem to biology influence, the sub-network 1 by
CapsNet is constituted；
3) original block is input in sub-network 2, the inter-species for simulating biology influences, and the sub-network 2 is by LSTM and mutually
Attention mechanism module is constituted；It specifically includes:
(1) original block is input in LSTM, the output result of LSTM is input to mutual attention mechanism module, it is described
Mutual attention mechanism module is as shown in Fig. 2, include mutual attention mechanism submodule 1 and mutual attention mechanism submodule 2, wherein
The mutual attention mechanism submodule 1 is made of m × n full articulamentums and Softmax layers, the mutual attention mechanism
Submodule 2 is made of m × n mutual attention weight modules, and the mutual attention weight module is by every kind of biological attribute
M × n-1 weight vectors of vector form, and the element inside each weight vectors is identical；
(2) in mutual attention mechanism submodule 1, by each biological species attribute vector be separately input to full articulamentum and
It is calculated in Softmax layers, then the calculated result of mutual attention mechanism submodule 1 is input to mutual attention mechanism submodule
In 2, every kind of biological attribute vector can distribute m × n-1 weight vectors, in mutual attention mechanism submodule 2, by each biology
Attribute vector respectively with distribute to this kind of m × n-1 biological weight vectors corresponding element and be multiplied, and m × n-1 calculating is tied
Fruit is multiplied with the attribute vector of other biological respectively, each biology obtain other all biological m × n-1 weighting attributes to
Amount, and weight attribute vector for m × n-1 and be added the attribute vector new as this kind of biology；
(3) the output result of mutual attention mechanism module and the cosine similarity of tag block are calculated, similarity matrix is obtained
2, the output result of LSTM is cyclically input in LSTM, is trained and by the constantly readjustment LSTM of loss function 2 and mutually
The parameter of attention mechanism module.
4) initial sub-block 2 is input in sub-network 3, simulating in the kind of biology influences, the sub-network 3 be by LSTM and
It is constituted from attention mechanism module；It specifically includes:
(1) initial sub-block 2 is input in LSTM, the output result of LSTM is input to from attention mechanism module, certainly
Attention mechanism module is as shown in figure 3, comprising from attention mechanism submodule 1 and from attention mechanism submodule 2；It is described from
Attention mechanism submodule 1 is made of m × n full articulamentums and Softmax layers, described from attention mechanism submodule 2
It is made of m × n weight vectors, the element inside each weight vectors is identical；
(2) from attention mechanism submodule 1, by each biological species attribute vector be separately input to full articulamentum and
It is calculated, then will be input to from the calculated result of attention mechanism submodule 1 from attention mechanism submodule in Softmax layers
In 2, from attention mechanism submodule 2, by each biological attribute vector and this kind of biological 1 weight vectors pair are distributed to
It answers element multiplication, and calculated result is multiplied the attribute vector new as this kind of biology with the attribute vector of the biology；
(3) cosine similarity for calculating the output result and label sub-block 2 from attention mechanism module, obtains similarity moment
Battle array 3, the output result of LSTM is cyclically input in LSTM and is trained, and by loss function 3 constantly readjustment LSTM and
From the parameter of attention mechanism module.
5) similarity matrix 1, similarity matrix 2 and similarity matrix 3 are averaging by corresponding element, are obtained averagely similar
Matrix is spent, then all elements of average similarity matrix are averaging, to obtain similarity score.
Claims (6)
1. a kind of analogy method of bioecosystem neural network based, which comprises the steps of:
1) data set of the analogy method of bioecosystem neural network based is established, comprising: every kind of biological category of setting
Property, the coding of attribute vector and division simultaneously constitute tag block, bamboo slips used for divination or drawing lots block 1, label sub-block 2, original block, initial sub-block 1 and initial
Sub-block 2；
2) initial sub-block 1 is input in sub-network 1, model ecosystem to biology influence, the sub-network 1 by
CapsNet is constituted；
3) original block is input in sub-network 2, the inter-species for simulating biology influences, and the sub-network 2 pays attention to by LSTM and mutually
Power mechanism module is constituted；
4) initial sub-block 2 is input in sub-network 3, simulating in the kind of biology influences, and the sub-network 3 is to infuse by LSTM and certainly
Power mechanism module of anticipating is constituted；
5) similarity matrix 1, similarity matrix 2 and similarity matrix 3 are averaging by corresponding element, obtain average similarity square
Battle array, then all elements of average similarity matrix are averaging, to obtain similarity score.
2. a kind of analogy method of bioecosystem neural network based according to claim 1, which is characterized in that
Every kind of biological attribute described in step 1), comprising: boundary a, door b, guiding principle c, mesh d, section e, belong to f, kind g, ecosystem type h, cross
Across the ecosystem quantity i, population quantity j, quality k, natural enemy biological species number l, natural enemy biomass m, predation biological species number
N, biological total amount o is preyed on.
3. a kind of analogy method of bioecosystem neural network based according to claim 1, which is characterized in that
The coding of attribute vector described in step 1) and division simultaneously constitute tag block, bamboo slips used for divination or drawing lots block 1, label sub-block 2, original block, initial son
Block 1 and initial sub-block 2, comprising:
(1) by every kind of biological combinations of attributes at an attribute vector A with 15 attributes1=a, b, c, d, e, f, g, h, i,
J, k encode boundary a, door b, guiding principle c, mesh d, section e, category f, kind g using the mode of binary tree step by step；By attribute vector A1It draws respectively
It is divided into the attribute subvector A with 9 attributes2With the attribute subvector A with 6 attributes3, wherein A2=a, b, c, d, e, f,
G, h, i } and A3={ j, k, l, m, n, o }；
(2) attribute vector of m × n kind biology is arranged in the tensor of m × n × 15, constitutes tag block；By the son of m × n kind biology
Vector A2And A3It is arranged together in the tensor of m × n × 9 and m × n × 6, respectively constitutes label sub-block 1 and label sub-block 2；
(3) tensor of m × n × 15 for being arranged in m × n identical monadic attribute vectors constitutes original block；By m
× n identical the tensors of monadic m × n × 15 are divided into the tensor of m × n × 9 and m × n × 6, respectively constitute initial
Sub-block 1 and initial sub-block 2.
4. a kind of analogy method of bioecosystem neural network based according to claim 1, which is characterized in that
Step 2) includes: that the output result and 1 corresponding position attribute vector of label sub-block of CapsNet are calculated separately according to the type of biology
Cosine similarity, obtain similarity matrix 1；The output result of CapsNet is cyclically input in CapsNet and is trained
And pass through the constantly parameter of readjustment CapsNet of loss function 1.
5. a kind of analogy method of bioecosystem neural network based according to claim 1, which is characterized in that
Step 3) includes:
(1) original block is input in LSTM, the output result of LSTM is input to mutual attention mechanism module, the mutual note
Power mechanism module of anticipating includes mutual attention mechanism submodule 1 and mutual attention mechanism submodule 2, wherein the mutual attention
Mechanism submodule 1 is made of the full articulamentums of m × n and Softmax layers, the mutual attention mechanism submodule 2 be by m ×
N mutually attention weight module composition, the mutual attention weight module are by m × n-1 of every kind of biological attribute vector
Weight vectors form, and the element inside each weight vectors is identical；
(2) in mutual attention mechanism submodule 1, by each biological species attribute vector be separately input to full articulamentum and
It is calculated in Softmax layers, then the calculated result of mutual attention mechanism submodule 1 is input to mutual attention mechanism submodule
In 2, every kind of biological attribute vector can distribute m × n-1 weight vectors, in mutual attention mechanism submodule 2, by each biology
Attribute vector respectively with distribute to this kind of m × n-1 biological weight vectors corresponding element and be multiplied, and m × n-1 calculating is tied
Fruit is multiplied with the attribute vector of other biological respectively, each biology obtain other all biological m × n-1 weighting attributes to
Amount, and weight attribute vector for m × n-1 and be added the attribute vector new as this kind of biology；
(3) the output result of mutual attention mechanism module and the cosine similarity of tag block are calculated, similarity matrix 2 is obtained, it will
The output result of LSTM is cyclically input in LSTM, is trained and is paid attention to by the constantly readjustment LSTM of loss function 2 with mutual
The parameter of power mechanism module.
6. a kind of analogy method of bioecosystem neural network based according to claim 1, which is characterized in that
Step 4) includes:
(1) initial sub-block 2 is input in LSTM, the output result of LSTM is input to from attention mechanism module, from attention
Power mechanism module includes from attention mechanism submodule 1 and from attention mechanism submodule 2；It is described from attention mechanism submodule
Block 1 is made of the full articulamentums of m × n and Softmax layers, it is described from attention mechanism submodule 2 be by m × n weight
Vector forms, and the element inside each weight vectors is identical；
(2) from attention mechanism submodule 1, by each biological species attribute vector be separately input to full articulamentum and
It is calculated, then will be input to from the calculated result of attention mechanism submodule 1 from attention mechanism submodule in Softmax layers
In 2, from attention mechanism submodule 2, by each biological attribute vector and this kind of biological 1 weight vectors pair are distributed to
It answers element multiplication, and calculated result is multiplied the attribute vector new as this kind of biology with the attribute vector of the biology；
(3) cosine similarity for calculating the output result and label sub-block 2 from attention mechanism module, obtains similarity matrix 3,
The output result of LSTM is cyclically input in LSTM and is trained, and is infused by the constantly readjustment LSTM of loss function 3 and certainly
The parameter of meaning power mechanism module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810843096.6A CN109063319B (en) | 2018-07-27 | 2018-07-27 | Simulation method of biological ecosystem based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810843096.6A CN109063319B (en) | 2018-07-27 | 2018-07-27 | Simulation method of biological ecosystem based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109063319A true CN109063319A (en) | 2018-12-21 |
| CN109063319B CN109063319B (en) | 2023-04-07 |
Family
ID=64836826
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810843096.6A Active CN109063319B (en) | 2018-07-27 | 2018-07-27 | Simulation method of biological ecosystem based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109063319B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102496077A (en) * | 2011-12-06 | 2012-06-13 | 北京林业大学 | System and method for predicting forest pest disaster |
| US20140029196A1 (en) * | 2011-04-15 | 2014-01-30 | Kevin Smith | System for balanced power and thermal management of mission critical environments |
| CN107392224A (en) * | 2017-06-12 | 2017-11-24 | 天津科技大学 | A kind of crop disease recognizer based on triple channel convolutional neural networks |
| CN107886073A (en) * | 2017-11-10 | 2018-04-06 | 重庆邮电大学 | A kind of more attribute recognition approaches of fine granularity vehicle based on convolutional neural networks |
| CN108171322A (en) * | 2017-12-25 | 2018-06-15 | 青岛理工大学 | A kind of Learning Algorithm based on particle group optimizing |
| CN108319928A (en) * | 2018-02-28 | 2018-07-24 | 天津大学 | A kind of deep learning model and application based on Multi-objective PSO optimization |
-
2018
- 2018-07-27 CN CN201810843096.6A patent/CN109063319B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140029196A1 (en) * | 2011-04-15 | 2014-01-30 | Kevin Smith | System for balanced power and thermal management of mission critical environments |
| CN102496077A (en) * | 2011-12-06 | 2012-06-13 | 北京林业大学 | System and method for predicting forest pest disaster |
| CN107392224A (en) * | 2017-06-12 | 2017-11-24 | 天津科技大学 | A kind of crop disease recognizer based on triple channel convolutional neural networks |
| CN107886073A (en) * | 2017-11-10 | 2018-04-06 | 重庆邮电大学 | A kind of more attribute recognition approaches of fine granularity vehicle based on convolutional neural networks |
| CN108171322A (en) * | 2017-12-25 | 2018-06-15 | 青岛理工大学 | A kind of Learning Algorithm based on particle group optimizing |
| CN108319928A (en) * | 2018-02-28 | 2018-07-24 | 天津大学 | A kind of deep learning model and application based on Multi-objective PSO optimization |
Non-Patent Citations (1)
| Title |
|---|
| 张力等: "基于GA-BP人工神经网络的樟树林土壤呼吸对施氮响应的研究", 《生态学报》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109063319B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Salmeron et al. | Learning fuzzy cognitive maps with modified asexual reproduction optimisation algorithm | |
| Han et al. | A survey on metaheuristic optimization for random single-hidden layer feedforward neural network | |
| Zhang et al. | ANODEV2: A coupled neural ODE framework | |
| CN109657156B (en) | Individualized recommendation method based on loop generation countermeasure network | |
| CN108334605B (en) | Text classification method and device, computer equipment and storage medium | |
| CN109948029A (en) | Based on the adaptive depth hashing image searching method of neural network | |
| CN110188358A (en) | The training method and device of Natural Language Processing Models | |
| CN110472730A (en) | A kind of distillation training method and the scalable dynamic prediction method certainly of convolutional neural networks | |
| CN108647251A (en) | The recommendation sort method of conjunctive model is recycled based on wide depth door | |
| Wang et al. | A hybrid GA-PSO method for evolving architecture and short connections of deep convolutional neural networks | |
| Bhandare et al. | Designing convolutional neural network architecture using genetic algorithms | |
| CN108596327A (en) | A kind of seismic velocity spectrum artificial intelligence pick-up method based on deep learning | |
| CN106503723A (en) | A kind of video classification methods and device | |
| CN107169574A (en) | Using nested machine learning model come the method and system of perform prediction | |
| CN108416460A (en) | Cyanobacterial bloom prediction technique based on the random depth confidence network model of multifactor sequential- | |
| Chen et al. | Binarized neural architecture search for efficient object recognition | |
| Bakhshi et al. | Fast evolution of CNN architecture for image classification | |
| Sadr et al. | Convolutional neural network equipped with attention mechanism and transfer learning for enhancing performance of sentiment analysis | |
| CN109889923A (en) | Utilize the method for combining the layering of video presentation to summarize video from attention network | |
| CN112069399A (en) | Personalized search system based on interactive matching | |
| CN113011487B (en) | Open set image classification method based on joint learning and knowledge migration | |
| Strumberger et al. | Hybridized monarch butterfly algorithm for global optimization problems | |
| CN110289987A (en) | Multi-agent system network resilience appraisal procedure based on representative learning | |
| Czajkowski et al. | Steering the interpretability of decision trees using lasso regression-an evolutionary perspective | |
| CN112200208B (en) | Cloud workflow task execution time prediction method based on multi-dimensional feature fusion |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |