CN109598336A - A kind of Data Reduction method encoding neural network certainly based on stack noise reduction - Google Patents
A kind of Data Reduction method encoding neural network certainly based on stack noise reduction Download PDFInfo
- Publication number
- CN109598336A CN109598336A CN201811476554.3A CN201811476554A CN109598336A CN 109598336 A CN109598336 A CN 109598336A CN 201811476554 A CN201811476554 A CN 201811476554A CN 109598336 A CN109598336 A CN 109598336A
- Authority
- CN
- China
- Prior art keywords
- layer
- dae
- reduction
- neural network
- noise reduction
- 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
- 230000009467 reduction Effects 0.000 title claims abstract description 90
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 35
- 230000008569 process Effects 0.000 claims abstract description 16
- 238000010276 construction Methods 0.000 claims abstract description 4
- 210000002569 neuron Anatomy 0.000 claims description 22
- 230000007423 decrease Effects 0.000 claims description 3
- 230000000977 initiatory effect Effects 0.000 claims description 3
- 238000010801 machine learning Methods 0.000 abstract description 4
- 230000006870 function Effects 0.000 abstract description 3
- 238000002474 experimental method Methods 0.000 description 8
- 238000012360 testing method Methods 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 210000005036 nerve Anatomy 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Compression Of Band Width Or Redundancy In Fax (AREA)
Abstract
The invention discloses a kind of Data Reduction methods for encoding neural network certainly based on stack noise reduction, it is characterized in that, the reduction model construction step of stack noise reduction from coding neural network is as follows: step 1: by the output of previous DAE, as the input of next DAE, achieve the purpose that encode layer by layer with this;Step 2: usingIt indicates original input sample, is used in combinationCome represent i-th layer DAE coding situation, it can be deduced that the coding situation of each layer of DAE;Successively greedy training and fine tuning are carried out, trim process adjusts the intersection entropy function of initial parameter by BP algorithm to guarantee the minimum of reconstructed error.The present invention is using improved method-stack noise reduction of noise reduction autoencoder network from neural network algorithm is encoded to sample characteristics collection progress dimensionality reduction, to reduce the complexity of each class model, the classifying quality of classifier in raising and machine learning application, the operation cost of all kinds of learning algorithms is reduced, and the feasibility and high efficiency of this method reduction are verified.
Description
Technical field
The present invention relates to technical field of data processing, in particular to a kind of number for encoding neural network certainly based on stack noise reduction
According to reduction method.
Background technique
Self-encoding encoder (Autoencoder, AE) is to propose that structure was divided into input layer, defeated in 2006 by Hinton
Layer and its hidden layer out.Input layer is identical with output layer neuron quantity, hidden layer neuron negligible amounts, wherein input layer
Coding network part is constituted with hidden layer, data are partially compressed in coding network in AE.
Currently, the problem of data redundancy, is increasingly severe, it is not only with data acquisition, the fast development of memory technology
Memory space is greatly wasted, the modeling based on data can be also significantly reduced.
Big, the stronger feature of relevance between index for the dimension height of mass data collection, redundancy, to improve to data
Reason ability and availability of data need to propose that a kind of novel method pre-processes initial data.
Summary of the invention
Invention is designed to provide a kind of Data Reduction method based on stack noise reduction from coding neural network, the present invention
Dimensionality reduction is carried out to sample characteristics collection from coding neural network algorithm using improved method-stack noise reduction of noise reduction autoencoder network,
To reduce the complexity of each class model, the classifying quality of classifier, reduces all kinds of learning algorithms in raising and machine learning application
Operation cost, and the feasibility and high efficiency of this method reduction are verified, to solve to propose in above-mentioned background technique
The problem of.
To achieve the above object, the invention provides the following technical scheme:
A kind of Data Reduction method encoding neural network certainly based on stack noise reduction, passes through qD, initial data X multilated
At, and using this with noisy data as the input of self-encoding encoder, pass through fθ, to the activation value of each neuron of hidden layer into
Row calculates, and the reduction model construction step of stack noise reduction from coding neural network is as follows:
Step 1: the output of previous DAE as the input of next DAE, achievees the purpose that encode layer by layer with this;
Step 2: using x0It indicates original input sample, and uses xiCome represent i-th layer DAE coding situation, can obtain
The coding situation of each layer of DAE, representation are as follows out:
xi=fθ(xi-1) i=1,2,3 ....
Step 3: carrying out successively greedy training and fine tuning, wherein successively greediness training process is originally inputted by minimizing
The difference training weight of data and reconstruct coding, obtains initial parameter, trim process adjusts the friendship of initial parameter by BP algorithm
Entropy function is pitched to guarantee the minimum of reconstructed error, to obtain the purpose of optimal quality reconstruction.
Further, it when the SDAE that training is made of multilayer DAE, needs using successively greedy principle, to each layer of DAE
It carries out individually training and obtains initiation parameter, and parameter is finely adjusted on the basis of guaranteeing that reconstructed error minimizes.
Further comprising following steps: first with input sample characteristics training SDAE first layer, i.e., first
DAE, and corresponding parameter is obtained by fine tuning, the hidden layer of the DAE is then exported into the input as second DAE, training
And finely tune and obtain the parameter of second DAE, successively go down, the reduction model based on SDAE can be obtained.
Further, in entire training process, to guarantee the parameter constant of a upper DAE when training next DAE.
It further, further include the spam page discriminant criterion reduction being made of input layer, hidden layer and output layer
The network structure of model, the structure of every layer of DAE are respectively 219-150,150-100,100-50,50-5, wherein setting input
The neuron number of layer is 219, and the neuron number that output layer is arranged is 5, meanwhile, to achieve the purpose that reduction, it is arranged every layer
150,100,50 layer-by-layer decline trend is presented in the number of neuron.
Further, in reduction model, the hidden layer output of every layer of DAE is respectively the input of next layer of DAE, by layer-by-layer
Between study so that the neuron of next layer of DAE can capture the high correlation of the neuron of preceding layer DAE, and can be accurate
The non-linear relation for describing the neuron of preceding layer DAE, enables final exports coding to be fully contemplated by the information of high dimensional data.
Compared with prior art, the beneficial effects of the present invention are: the stack noise reduction proposed by the present invention that is based on is from encoding nerve
The Data Reduction method of network first analyzes data set progress in detail comprehensively, is quantified to sample data set, standardized
And equilibrating processing, using improved method-stack noise reduction of noise reduction autoencoder network from neural network algorithm is encoded to sample spy
Collection carries out dimensionality reduction, and to reduce the complexity of each class model, the classifying quality of classifier, is reduced in raising and machine learning application
The operation cost of all kinds of learning algorithms, and the feasibility and high efficiency of this method reduction are verified.
Detailed description of the invention
Fig. 1 is the schematic diagram of DAE of the invention;
Fig. 2 is the network structure of first DAE of the invention;
Fig. 3 is the network structure of second DAE of the invention;
Fig. 4 is the reduction model figure of the invention based on SDAE;
Fig. 5 is the network structure of spam page discriminant criterion reduction model of the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
A kind of Data Reduction method encoding neural network certainly based on stack noise reduction, on the basis of the characteristics of retaining AE,
Noise reduction self-encoding encoder (Denoising Autoencoder, DAE) learns AE from noise-containing input, by
Certain noises are added in input data, the robustness of Lai Tigao system, the schematic diagram of DAE is as shown in Figure 1, pass through qD, original number
According to X multilated at, and using this with noisy data as the input of self-encoding encoder, pass through fθ, to each neuron of hidden layer
Activation value calculated, stack noise reduction from coding neural network reduction model construction step it is as follows:
Step 1: encoding neural network (Stacked Denoising Autoencoder Neural certainly in stack noise reduction
Networks, SDAE) in model, the output of previous DAE is encoded with this to reach layer by layer as the input of next DAE
Purpose;
Step 2: using x0It indicates original input sample, and uses xiCome represent i-th layer DAE coding situation, can obtain
The coding situation of each layer of DAE, representation are as follows out:
xi=fθ(xi-1) i=1,2,3 ....
Step 3: successively greedy training and fine tuning are carried out during constructing SDAE reduction model, wherein successively greedy
Training process obtains initial parameter, trim process by minimizing original input data and reconstructing the difference training weight of coding
The intersection entropy function of initial parameter is adjusted to guarantee the minimum of reconstructed error, to obtain optimal quality reconstruction by BP algorithm
Purpose.
When the SDAE that training is made of multilayer DAE, need to carry out each layer of DAE independent using successively greedy principle
Training obtains initiation parameter, and is finely adjusted on the basis of guaranteeing that reconstructed error minimizes to parameter.I.e. first with input
Sample characteristics training SDAE first layer, i.e. first DAE, process such as Fig. 2, and by fine tuning obtain corresponding parameter, so
The hidden layer of the DAE is exported into the input as second DAE, process such as Fig. 3 afterwards, training simultaneously finely tunes to obtain second DAE's
Parameter is successively gone down, and the reduction model based on SDAE, process such as Fig. 4 can be obtained.In entire training process, under training
To guarantee the parameter constant of a upper DAE when one DAE.
It is 219 dimensions for spam page discriminant criterion sample dimension, the present invention chooses the stack noise reduction with 4 layers of DAE certainly
Encoding nerve network structure comprising input layer, hidden layer and output layer, the structure of every layer of DAE be respectively 219-150,
150-100,100-50,50-5, wherein the neuron number that input layer is arranged is 219, and the neuron number that output layer is arranged is
5, meanwhile, to achieve the purpose that reduction, 150,100,50 layer-by-layer decline trend is presented in the number that every layer of neuron is arranged.Reduction
In model, the hidden layer output of every layer of DAE is respectively the input of next layer of DAE, by the study between successively, so that next layer
The neuron of DAE can capture the high correlation of the neuron of preceding layer DAE, and can accurate description preceding layer DAE nerve
The non-linear relation of member, enables final exports coding to be fully contemplated by the information of high dimensional data, the spam page discriminant criterion
The network structure of reduction model is illustrated in fig. 5 shown below.
By the foundation of above Indexes Reduction model carry out Indexes Reduction experimental result and analysis, haphazard selection 120
A sample pre-processes this 120 experiment samples with preprocess method, and using wherein 3/4 experiment sample as instruction
Practice collection, 1/4 experiment sample is as test set.
The characteristics for being tieed up the spam page discriminant criterion sample reduction of 219 dimensions for 5 with obtained SDAE model, point
Not Xuan Qu a training sample and test sample carry out reduction experiment, the characteristic of 5 dimensions obtained after reduction is such as
Shown in table 1.
Characteristic after 1 reduction of table
Original data can be completely represented for the data after verifying reduction, spam page reduction will be used for from SDAE model
Feasibility, and the aspect of validity two of work, which is verified, to be differentiated to subsequent spam page to this kind of reduction model.
Carry out feasibility analysis, verifying SDAE is used for the feasibility of reduction, be substantially verify the obtained coding of reduction can
Represent the information that original index is included, i.e. data after verifying reconstruct and the difference between legacy data.
(1) evaluation criterion
This experiment chooses mean square error (Mean Squared Error, MSE) as reconstructed error in measurement training process
Evaluation criterion.So, shown in MSE is defined as follows:
In formula, training sample or test sample ydataIt indicates, the sample y reconstructedreconIndicate, training sample or
The quantity of test sample is indicated with N.
(2) experimental result and analysis
Using treated, 219 dimension samples are used as the input of SDAE, and using 5 dimension codings as exporting, the present invention devises 4 layers
SDAE network structure, DAE1 219-150, DAE2 150-100, DAE3 100-50, DAE4 50-5, each layer neuronal quantity
Are as follows: 219-150-100-50-5.The frequency of training that each layer DAE and SDAE model is arranged is 10 times.
During each layer DAE training 10 times, by continuous adjusting parameter, the change of the reconstructed error of obtained each layer DAE
Change curve, when frequency of training reaches 5 times, the MSE of each layer DAE basically reaches stable state, that is, work as instruction already less than 0.009
When white silk number is set greater than equal to 5, the initial parameter of each layer can be obtained, the present invention will train 10 obtained each layer DAE's
Initial parameter of the parameter as each layer DAE of SDAE model.And the model is used to verify stack noise reduction and is used from coding neural network
In the feasibility of reduction.
During SDAE is trained to 10 times, the change curve of obtained reconstructed error, after 10 training, MSE is missed
Difference basically reaches stabilization, according to Matlab experimental result it is found that the minimal reconstruction error that sample data reaches less than 0.0024
It is 0.00226.That is by training, the coding of 5 dimensions can map the original number of original 219 dimension with minimum reconstructed error
According to.
Carry out high efficiency analysis, in order to prove the high efficiency of SDAE, the present invention will compare disaggregated model based on SDAE with
The reduction effect of other disaggregated models.
(1) evaluation criterion
By the evaluation criterion used for accuracy (Accuracy), i.e., the test sample number correctly classified accounts for survey for this experiment
The ratio of total sample number is tried, calculation formula is as follows:
Accuracy=(TN+TP)/C
Wherein, TN indicates that, by the spam page number of mistake classification, TP indicates the spam page number correctly classified, C
Indicate the sum of test sample.
(2) experimental result and analysis
SVM classifier influence to test result of the present invention by comparison based on SDAE, PCA, EMD is based on to verify
SDAE is used for the high efficiency of spam page reduction.Based on the SVM classifier of SDAE with the increase of experiment number, classification accuracy rate
In incremental trend, and compared with other two kinds of classifiers for, show higher recognition capability.
In summary two kinds of experiments are demonstrated by outputting and inputting the reconstructed error of data in comparison SDAE model
SDAE is used for the feasibility of reduction.By by SDAE and svm classifier models coupling composition and classification device, and with other two kinds of reduction sides
Method compares final result with the SVM classifier constituted, demonstrates the high efficiency of SDAE reduction method.
In conclusion the Data Reduction method proposed by the present invention for encoding neural network certainly based on stack noise reduction, right first
Data set progress analyze in detail comprehensively, sample data set is quantified, standardize and equilibrating handle, it is self-editing using noise reduction
Improved method-stack noise reduction of code network carries out dimensionality reduction to sample characteristics collection from coding neural network algorithm, to reduce all kinds of moulds
The classifying quality of classifier, reduces the operation cost of all kinds of learning algorithms in the complexity of type, raising and machine learning application, and
The feasibility and high efficiency of this method reduction are verified.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
Anyone skilled in the art within the technical scope of the present disclosure, according to the technique and scheme of the present invention and its
Inventive concept is subject to equivalent substitution or change, should be covered by the protection scope of the present invention.
Claims (6)
1. a kind of Data Reduction method for encoding neural network certainly based on stack noise reduction, which is characterized in that pass through qD, initial data X
Multilated at, and using this with noisy data as the input of self-encoding encoder, pass through fθ, each neuron of hidden layer is swashed
Value living is calculated, and the reduction model construction step of stack noise reduction from coding neural network is as follows:
Step 1: the output of previous DAE as the input of next DAE, achievees the purpose that encode layer by layer with this;
Step 2: using x0It indicates original input sample, and uses xiCome represent i-th layer DAE coding situation, it can be deduced that it is every
The coding situation of one layer of DAE, representation are as follows:
xi=fθ(xi-1) i=1,2,3 ....
Step 3: carrying out successively greedy training and fine tuning, wherein successively greediness training process is by minimizing original input data
With the difference training weight of reconstruct coding, initial parameter is obtained, trim process adjusts the cross entropy of initial parameter by BP algorithm
Function is to guarantee the minimum of reconstructed error, to obtain the purpose of optimal quality reconstruction.
2. a kind of Data Reduction method for encoding neural network certainly based on stack noise reduction according to claim 1, feature
It is, when training the SDAE being made of multilayer DAE, needs individually to train each layer of DAE using successively greedy principle
Initiation parameter is obtained, and parameter is finely adjusted on the basis of guaranteeing that reconstructed error minimizes.
3. a kind of Data Reduction method for encoding neural network certainly based on stack noise reduction according to claim 2, feature
It is comprising following steps: first with the first layer of the sample characteristics training SDAE of input, i.e. first DAE, and passes through micro-
Tune obtains corresponding parameter, and the hidden layer of the DAE is then exported the input as second DAE, trained and finely tune to obtain the
The parameter of two DAE, successively goes down, and can obtain the reduction model based on SDAE.
4. a kind of Data Reduction method for encoding neural network certainly based on stack noise reduction according to claim 3, feature
It is, in entire training process, guarantees the parameter constant of a upper DAE when training next DAE.
5. a kind of Data Reduction method for encoding neural network certainly based on stack noise reduction according to claim 1, feature
It is, further includes the network knot for the spam page discriminant criterion reduction model being made of input layer, hidden layer and output layer
Structure, the structure of every layer of DAE are respectively 219-150,150-100,100-50,50-5, wherein the neuron of input layer is arranged
Number is 219, and the neuron number that output layer is arranged is 5, meanwhile, to achieve the purpose that reduction, the number of every layer of neuron is set
150,100,50 layer-by-layer decline trend is presented.
6. a kind of Data Reduction method for encoding neural network certainly based on stack noise reduction according to claim 5, feature
It is, in reduction model, the hidden layer output of every layer of DAE is respectively the input of next layer of DAE, by the study between successively, is made
Next layer of DAE neuron can capture preceding layer DAE neuron high correlation, and can accurate description preceding layer
The non-linear relation of the neuron of DAE enables final exports coding to be fully contemplated by the information of high dimensional data.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811476554.3A CN109598336A (en) | 2018-12-05 | 2018-12-05 | A kind of Data Reduction method encoding neural network certainly based on stack noise reduction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811476554.3A CN109598336A (en) | 2018-12-05 | 2018-12-05 | A kind of Data Reduction method encoding neural network certainly based on stack noise reduction |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109598336A true CN109598336A (en) | 2019-04-09 |
Family
ID=65961118
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811476554.3A Pending CN109598336A (en) | 2018-12-05 | 2018-12-05 | A kind of Data Reduction method encoding neural network certainly based on stack noise reduction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109598336A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110411566A (en) * | 2019-08-01 | 2019-11-05 | 四川长虹电器股份有限公司 | A kind of Intelligent light spectrum signal denoising method |
CN110412872A (en) * | 2019-07-11 | 2019-11-05 | 中国石油大学(北京) | Reciprocating compressor fault diagnosis optimization method and device |
CN110866604A (en) * | 2019-10-28 | 2020-03-06 | 国网河北省电力有限公司电力科学研究院 | Cleaning method for power transformer state monitoring data |
CN111275074A (en) * | 2020-01-07 | 2020-06-12 | 东北电力大学 | Power CPS information attack identification method based on stack type self-coding network model |
CN112541874A (en) * | 2020-12-11 | 2021-03-23 | 福州大学 | Unsupervised denoising feature learning method based on self-encoder |
CN113220876A (en) * | 2021-04-16 | 2021-08-06 | 山东师范大学 | Multi-label classification method and system for English text |
CN113420794A (en) * | 2021-06-04 | 2021-09-21 | 中南民族大学 | Binaryzation Faster R-CNN citrus disease and pest identification method based on deep learning |
CN114689700A (en) * | 2022-04-14 | 2022-07-01 | 电子科技大学 | Low-power EMAT signal noise reduction method based on stack-type self-encoder |
CN115169499A (en) * | 2022-08-03 | 2022-10-11 | 中国电子科技集团公司信息科学研究院 | Asset data dimension reduction method and device, electronic equipment and computer storage medium |
CN116108386A (en) * | 2023-04-10 | 2023-05-12 | 南京信息工程大学 | Antique glass classification method and system under improved mixed sampling and noise reduction self-coding |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102750345A (en) * | 2012-06-07 | 2012-10-24 | 山东师范大学 | Method for identifying web spam through web page multi-view data association combination |
CN105163121A (en) * | 2015-08-24 | 2015-12-16 | 西安电子科技大学 | Large-compression-ratio satellite remote sensing image compression method based on deep self-encoding network |
CN106443447A (en) * | 2016-09-26 | 2017-02-22 | 南京航空航天大学 | An aero-generator fault feature extraction method based on iSDAE |
CN106874952A (en) * | 2017-02-16 | 2017-06-20 | 中国人民解放军国防科学技术大学 | Feature fusion based on stack self-encoding encoder |
CN107679031A (en) * | 2017-09-04 | 2018-02-09 | 昆明理工大学 | Based on the advertisement blog article recognition methods for stacking the self-editing ink recorder of noise reduction |
CN107749757A (en) * | 2017-10-18 | 2018-03-02 | 广东电网有限责任公司电力科学研究院 | A kind of data compression method and device based on stacking-type own coding and PSO algorithms |
CN107909105A (en) * | 2017-11-13 | 2018-04-13 | 上海交通大学 | A kind of Market Site Selection method and system |
CN108304359A (en) * | 2018-02-06 | 2018-07-20 | 中国传媒大学 | Unsupervised learning uniform characteristics extractor construction method |
-
2018
- 2018-12-05 CN CN201811476554.3A patent/CN109598336A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102750345A (en) * | 2012-06-07 | 2012-10-24 | 山东师范大学 | Method for identifying web spam through web page multi-view data association combination |
CN105163121A (en) * | 2015-08-24 | 2015-12-16 | 西安电子科技大学 | Large-compression-ratio satellite remote sensing image compression method based on deep self-encoding network |
CN106443447A (en) * | 2016-09-26 | 2017-02-22 | 南京航空航天大学 | An aero-generator fault feature extraction method based on iSDAE |
CN106874952A (en) * | 2017-02-16 | 2017-06-20 | 中国人民解放军国防科学技术大学 | Feature fusion based on stack self-encoding encoder |
CN107679031A (en) * | 2017-09-04 | 2018-02-09 | 昆明理工大学 | Based on the advertisement blog article recognition methods for stacking the self-editing ink recorder of noise reduction |
CN107749757A (en) * | 2017-10-18 | 2018-03-02 | 广东电网有限责任公司电力科学研究院 | A kind of data compression method and device based on stacking-type own coding and PSO algorithms |
CN107909105A (en) * | 2017-11-13 | 2018-04-13 | 上海交通大学 | A kind of Market Site Selection method and system |
CN108304359A (en) * | 2018-02-06 | 2018-07-20 | 中国传媒大学 | Unsupervised learning uniform characteristics extractor construction method |
Non-Patent Citations (4)
Title |
---|
吴德烽: "《计算智能及其在三维表面扫描机器人***中的应用》", 30 November 2012, 大连海事大学出版社 * |
张素智 等: "面向聚类的堆叠降噪自动编码器的特征提取研究", 《现代计算机》 * |
殷复莲: "《数据分析与数据挖掘实用教程》", 30 September 2017, 中国传媒大学出版社 * |
董海鹰: "《智能控制理论及应用》", 30 September 2016, 中国铁道出版社 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110412872A (en) * | 2019-07-11 | 2019-11-05 | 中国石油大学(北京) | Reciprocating compressor fault diagnosis optimization method and device |
CN110411566A (en) * | 2019-08-01 | 2019-11-05 | 四川长虹电器股份有限公司 | A kind of Intelligent light spectrum signal denoising method |
CN110866604A (en) * | 2019-10-28 | 2020-03-06 | 国网河北省电力有限公司电力科学研究院 | Cleaning method for power transformer state monitoring data |
CN111275074B (en) * | 2020-01-07 | 2022-08-05 | 东北电力大学 | Power CPS information attack identification method based on stacked self-coding network model |
CN111275074A (en) * | 2020-01-07 | 2020-06-12 | 东北电力大学 | Power CPS information attack identification method based on stack type self-coding network model |
CN112541874A (en) * | 2020-12-11 | 2021-03-23 | 福州大学 | Unsupervised denoising feature learning method based on self-encoder |
CN113220876A (en) * | 2021-04-16 | 2021-08-06 | 山东师范大学 | Multi-label classification method and system for English text |
CN113420794A (en) * | 2021-06-04 | 2021-09-21 | 中南民族大学 | Binaryzation Faster R-CNN citrus disease and pest identification method based on deep learning |
CN114689700A (en) * | 2022-04-14 | 2022-07-01 | 电子科技大学 | Low-power EMAT signal noise reduction method based on stack-type self-encoder |
CN114689700B (en) * | 2022-04-14 | 2023-06-06 | 电子科技大学 | Low-power EMAT signal noise reduction method based on stack-type self-encoder |
CN115169499A (en) * | 2022-08-03 | 2022-10-11 | 中国电子科技集团公司信息科学研究院 | Asset data dimension reduction method and device, electronic equipment and computer storage medium |
CN115169499B (en) * | 2022-08-03 | 2024-04-05 | 中国电子科技集团公司信息科学研究院 | Asset data dimension reduction method, device, electronic equipment and computer storage medium |
CN116108386A (en) * | 2023-04-10 | 2023-05-12 | 南京信息工程大学 | Antique glass classification method and system under improved mixed sampling and noise reduction self-coding |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109598336A (en) | A kind of Data Reduction method encoding neural network certainly based on stack noise reduction | |
Lin et al. | Audio classification and categorization based on wavelets and support vector machine | |
CN110930976B (en) | Voice generation method and device | |
CN111861945B (en) | Text-guided image restoration method and system | |
CN111652049A (en) | Face image processing model training method and device, electronic equipment and storage medium | |
CN109919364A (en) | Multivariate Time Series prediction technique based on adaptive noise reduction and integrated LSTM | |
CN110175560A (en) | A kind of radar signal intra-pulse modulation recognition methods | |
CN107220594B (en) | Face posture reconstruction and recognition method based on similarity-preserving stacked self-encoder | |
CN110269625A (en) | A kind of electrocardio authentication method and system of novel multiple features fusion | |
CN110726898B (en) | Power distribution network fault type identification method | |
CN113312989B (en) | Finger vein feature extraction network based on aggregated descriptors and attention | |
CN110379491A (en) | Identify glioma method, apparatus, equipment and storage medium | |
CN108805802A (en) | A kind of the front face reconstructing system and method for the stacking stepping self-encoding encoder based on constraints | |
Shankar et al. | Non-parallel emotion conversion using a deep-generative hybrid network and an adversarial pair discriminator | |
CN113886792A (en) | Application method and system of print control instrument combining voiceprint recognition and face recognition | |
CN113035228A (en) | Acoustic feature extraction method, device, equipment and storage medium | |
CN100369047C (en) | Image identifying method based on Gabor phase mode | |
CN112347910A (en) | Signal fingerprint identification method based on multi-mode deep learning | |
CN108831486B (en) | Speaker recognition method based on DNN and GMM models | |
Malik | Fighting AI with AI: fake speech detection using deep learning | |
CN116613740A (en) | Intelligent load prediction method based on transform and TCN combined model | |
CN115984911A (en) | Attribute generation countermeasure network and face image continuous transformation method based on same | |
CN111104868B (en) | Cross-quality face recognition method based on convolutional neural network characteristics | |
CN114283301A (en) | Self-adaptive medical image classification method and system based on Transformer | |
Al-Rawi et al. | Feature Extraction of Human Facail Expressions Using Haar Wavelet and Neural network |
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: 20190409 |
|
RJ01 | Rejection of invention patent application after publication |