CN107729942A - A kind of sorting technique of structured view missing data - Google Patents
A kind of sorting technique of structured view missing data Download PDFInfo
- Publication number
- CN107729942A CN107729942A CN201710990955.XA CN201710990955A CN107729942A CN 107729942 A CN107729942 A CN 107729942A CN 201710990955 A CN201710990955 A CN 201710990955A CN 107729942 A CN107729942 A CN 107729942A
- Authority
- CN
- China
- Prior art keywords
- mrow
- data
- view
- missing
- msubsup
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24143—Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of sorting technique of structured view missing data.Belong to data mining technology field.The framework has two stages:Structured view missing data processing stage, structured view missing multitask multi views sorting phase.1. structured view missing data processing stage:Build sample graph;Determine weight;Sample is changed to Feature Mapping space;Delete missing view.2. structured view lacks multitask multi views sorting phase:Select suitable grader;Train classifier parameters;The sample architecture after view will be deleted into complete data collection;Classified using the MMLE frameworks of proposition.The present invention can be used for the information of similar data in actual scene inconsistent or initial data preserve during the data gaps and omissions that occurs, initial data feature can farthest be extracted, the distribution of retention data and structure simultaneously, greatly improve the performance and application of the classification of multitask multi views.
Description
Technical field
The invention belongs to data mining technology field.
Background technology
With the arrival in big data epoch, the data volume increasingly accumulated and miscellaneous data structure allow researchers' mesh
It is too busy and connects.Wherein, the data of multitask multi views already become a reality one of most common data in application scenarios.Chinese songs
With the classification of English song, the division of different school curricula homepages, or even in medical domain, according to expert diagnosis and CT data etc.
Judgement to various disease, belong to multitask multi views field.Traditional is carried out to the data of single task role or single view
Analysis can not meet premise and the requirement of current data analysis.
The study of multitask multi views has turned into the important subject of data mining, to realize the difference disease in medical domain
Disease differentiates, the multiple sentiment analysis in social networks, and the student interests in education sector excavate etc., based on multitask multi views
Data analysis into main trend.
However, the complete data used in data analysis process is often the perfect condition that researcher contemplates, it is actual
It is usually not possible to get complete data, therefore, the analysis for missing data is imperative in scene.Existing missing data
Analysis, largely only considered the missing data how repaired in single view or single task role, then according to repairing after
As a result complete data is formed, so as to realize follow-up classification or cluster task.However, follow-up classification or cluster task often by
Data set has a great influence, and uncertain factor possessed by the data of repairing may result in classification accuracy reduction.So have
Necessity provides a kind of sorting technique of the structured view missing data based on multitask multi views, to the missing data of structuring
Angle consideration is changed, widens the research range of data analysis, improves the accuracy rate of data classification.
Found by the retrieval to existing patent and correlation technique, it is existing to be lacked with the view based on multitask multi views
The relevant method of class of losing points has:
(1) good, the clustering method of the attribute missing data collection of the such as Zhang Shengnan mono-, CN106127262A [P] .2016. are appointed
The clustering method of an attribute missing data collection is proposed, takes nearest neighbor method to determine the valuation constraint of missing attribute
Space, real coding is carried out to missing attribute and cluster centre, and optimizing is carried out by ant group algorithm, same during Optimized Iterative
When obtain missing attributes estimation value and cluster centre, then fuzzy clustering is completed by the membership function of FCM algorithms, it is basic herein
Upper formation attribute missing data collection mixes optimization clustering algorithm.
(2) Yuan Yubo, Ruan Tong, Qiu Wen wait by force a kind of missing data complementing methods based on K plane regressions of,
CN105469123A[P].2016.
Propose a kind of new method of database missing data completion.It is broadly divided into five steps:The first step, shortage of data
Detection, missing detection is carried out to providing data set;Second step is that the dimension of input variable about subtracts, the phase between analysis input dimension
Guan Xing, selected using pivoting (PCA) compared with correlated inputs dimension, form new input data set;3rd step is k points of training set
Cut, train set to split using cluster (Kmeans) input, obtained k classification training set;4th step, k planes
Regression function constructs, and obtains the geometric center of optimum regression coefficient and each plane, provides regression fit function;It is finally data
Completion is tested.
(3) thank by force, a kind of improvement missing datas based on KNN of the such as Wang Zhen fill up algorithm, CN106407464A [P]
.2017.
Propose a kind of improvement missing data based on KNN and fill up algorithm, it is first, reciprocal to traditional multiple correlation coefficient to assign
Power method is improved, and deletes some and less attribute is associated with determinant attribute, and property set is carried out to simplify operation, is obtained only containing essence
The set of data samples of simple property set;Then the advantages of correlation and variability between attribute are considered using mahalanobis distance, knot
Effective prediction of the gray relative analysis method to the sample containing uncertain factor is closed, calculates K neighbour's sample of missing sample;Finally
According to K distance value being calculated, according to entropy assessment to attribute tax entropy weight corresponding to K sample, in conjunction with property value,
Calculate and final fill up value.
(4) a kind of methods for missing data recover processing of the such as Guo Jinyu, Yuan Tangming, Li Yuan, CN104461772A
[P].2015.
Using KNN rules, by extracting complete data set in industry, the missing corresponding k neighbour of sample is calculated, is used
In the relevant information for being extracted in missing data in missing data sample local message.Application error minimizes criterion, calculates this and lacks
Lose the weight of neighbour's sample of data.Then reconstruct is weighted to corresponding neighbour's sample, reconstructs the data point of missing.
Although it can be seen that existing method has the advantages of certain but also there is some shortcomings:First, most of sides
Method all simply considers the situation that individual data lacks in single view, does not account for multiple characteristic strips possessed by multiple views
The influence come;Secondly, the processing method of individual data missing is typically to be filled up using the similar features with its neighbour, so
And the accuracy that the uncertainty and error filled up may result in final data classification reduces.
In order to overcome the weak point of existing method simultaneously, present invention employs Laplce's feature based on multi views to reflect
Shooting method, manifold structure and the distribution of multidimensional data are farthest retained with this;In addition, in order to improve classification performance with
Practical application efficiency, invention introduces the thought of multitask, can ensure that multiple tasks are carried out in a framework simultaneously, greatly
Efficiency of algorithm is improved greatly, reduces time complexity;It is contemplated that the effective information in mining data as much as possible, is carried
The feature for evidence of fetching, the performance of lifting data classification, generality of the raising for the method for missing data.Therefore, the present invention carries
The missing viewdata classification of the structuring based on multitask multi views gone out has higher Research Significance and application value.In view of
The deficiency of existing program set forth above, the present invention is intended to provide simpler, more perfect scheme, and be allowed to overcome existing skill
The disadvantage mentioned above of art.
The content of the invention
It is an object of the invention to provide a kind of sorting technique of structured view missing data, it can efficiently solve data
The problem of error that mending tape comes.
The technical solution adopted in the present invention is:
A kind of sorting technique of structured view missing data, it is a kind of MMLE based on multitask multi views
(Multi-task Multi-view Laplacian Eigenmaps) framework is used for the information content for enriching initial data, the frame
Frame includes two stages:The data processing of structured view missing, the multitask multi views classification of structured view missing.It is wrapped
Include following steps:
Step 1: the processing of structuring missing data, processing step are as follows:
(1) structure figure, is built into a figure by all points using KNN algorithms, K nearest point of each point is connected
Side.K is a value set in advance.Here without full-mesh method is used, because the dilute of data can be ensured using KNN algorithms
Dredge property.
(2) determine weight, it is determined that weight size between points, such as determined from heat kernel function, such as fruit dot i and
Point j is connected, then the weight setting of their relations is:
Wherein, XiAnd XjRespectively i-th and j-th of object;WijFor the weight between i-th of object and j-th of object;
E is then the natural constant in exponential function.
(3) Feature Mapping, the eigen vector of Laplce's eigenmatrix is calculated, Ly=λ Dy, wherein L are figures
Laplacian Matrix, λ are characteristic values, and y is characteristic vector, and D is diagonal matrix, meets Dii=∑jWji, L=D-W.If xiAfter mapping
Point be yi, then the object function of laplacian eigenmaps be:
(4) view of structural data missing is positioned, deletes the related value of the view, set to 0;Ignore for
The data of structured view missing, only consider that extraction does not lack the feature of view, and then complete lacking based on multitask multi views
Lose data classification;
Step 2: being classified using MMLE frameworks, its step is as follows:
(1) it is V={ v to define view1,v2,......vm, task is T={ t1,t2,......tn, the data after mapping
Put and beThen the object function of complete data is:
Wherein,WithI-th and j-th of data point under task t v-th of view are represented respectively;Represent
Weight in task t v-th of view between data point i and j.
(2) view of definition structure missing data is vl, in the case of structural data lacks, it is believed that the view
Data can not all be known, therefore directly leave out the value related to the view and set to 0, and renewal object function is:
(3) optimal classification device is selected, such as:KNN graders are classified to sample point.Obtain each sample point after projection
The distance between, it is determined that k neighbours of each data point, divide classification.The distance metric of sample point is as follows:
Wherein, l represents the dimension of sample data;P represents the norm of distance metric;Lp(yi,yj) represent sample yiWith yjIt
Between Lp distances.
It is that the parameter involved by Feature Mapping, the calculating of weight and selection of grader etc. can root actually implementing
It is altered according to the actual distribution situation of data.
Compared with prior art, advantages of the present invention and effect:
The problem of present invention is directed to shortage of data, first amplification have arrived the situation of structured view missing, specify that missing
The different situations of data classification, improve the accuracy of such method;Secondly, structured view Deletion Procedure can be in certain feature
The advantages of in the case of missing using multi views study, the feature of data is effectively excavated, lacking for information extraction can be filled up
Leakage;Finally, the concept of the multi-task learning of introducing can greatly promote classification effectiveness.Therefore, framework proposed by the invention tool
There is a high efficiency, low complex degree, the advantages that high applicability.
Brief description of the drawings
Fig. 1 is the flow chart of the present invention.
Embodiment
Implement to be described in further detail the present invention below in conjunction with accompanying drawing.
The present invention proposes a kind of MMLE (Multi-task Multi-view based on multitask multi views
Laplacian Eigenmaps) framework is used to enrich the information content of initial data, and the framework includes two stages:Structuring regards
Scheme the data processing of missing, the multitask multi views of structured view missing are classified, and it comprises the following steps:
Step 1: the processing of structuring missing data, as shown in figure 1, processing step is as follows:
(1) structure figure, is built into a figure by all points using KNN algorithms, K nearest point of each point is connected
Side.K is a value set in advance.Here without full-mesh method is used, because the dilute of data can be ensured using KNN algorithms
Dredge property.
(2) determine weight, it is determined that weight size between points, such as determined from heat kernel function, such as fruit dot i and
Point j is connected, then the weight setting of their relations is:
Wherein, XiAnd XjRespectively i-th and j-th of object;WijFor the weight between i-th of object and j-th of object;
E is then the natural constant in exponential function.
(3) Feature Mapping, the eigen vector of Laplce's eigenmatrix is calculated, Ly=λ Dy, wherein L are figures
Laplacian Matrix, λ are characteristic values, and y is characteristic vector, and D is diagonal matrix, meets Dii=∑jWji, L=D-W.If xiAfter mapping
Point be yi, then the object function of laplacian eigenmaps be:
(4) view of structural data missing is positioned, deletes the related value of the view, it is set to 0 one by one;Ignore
For the data of structured view missing, only consider that extraction does not lack the feature of view, and then complete to be based on multitask multi views
Missing data classification;
Step 2: being classified using MMLE frameworks, its step is as follows:
(1) it is V={ v to define view1,v2,......vm, task is T={ t1,t2,......tn, the data after mapping
Put and beThen the object function of complete data is:
Wherein,WithI-th and j-th of data point under task t v-th of view are represented respectively;Represent
Weight in task t v-th of view between data point i and j.
(2) view of definition structure missing data is vl, in the case of structural data lacks, it is considered herein that should
The data of view can not all be known, therefore directly leave out the value related to the view and set to 0, and renewal object function is:
(3) selection KNN graders are classified to sample point.The distance between each sample point after projection is obtained, really
K neighbours of fixed each data point, divide classification.The distance metric of sample point is as follows:
Wherein, l represents the dimension of sample data;P represents the norm of distance metric;Lp(yi,yj) represent sample yiWith yjIt
Between Lp distances.
It will be apparent to one skilled in the art that the invention is not restricted to the details of above-mentioned one exemplary embodiment, without departing substantially from
In the case of the spirit or essential attributes of the present invention, the present invention can be realized in other specific forms.The scope of the present invention by
Appended claims rather than described above limit, it is intended that will fall in the implication and scope of the equivalency of claim
All changes be included in the present invention.
Claims (1)
1. a kind of sorting technique of structured view missing data, comprises the following steps:
Step 1: the data processing of structured view missing:
Under same task between the data of different views figure structure, the determination of weight between all data points, all views
The laplacian eigenmaps of data sample, the view positioning of structural data missing, delete the correlation for lacking view;
(1) structure figure, is built into a figure by all points using KNN algorithms, K nearest point of each point is connected into side, K
It is a value set in advance;Here without full-mesh method is used, because the sparse of data can be ensured using KNN algorithms
Property;
(2) weight is determined, it is determined that weight size between points, is determined from heat kernel function, as fruit dot i is connected with point j,
The weight setting of so their relations is:
Wherein, XiAnd XjRespectively i-th and j-th of object;WijFor the weight between i-th of object and j-th of object;E is then
Natural constant in exponential function;
(3) Feature Mapping, the eigen vector of Laplce's eigenmatrix is calculated, Ly=λ Dy, wherein L are that Tula is general
Lars matrix, λ are characteristic values, and y is characteristic vector, and D is diagonal matrix, meets Dii=∑jWji, L=D-W;If xiPoint after mapping
For yi, then the object function of laplacian eigenmaps be:
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</munder>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<msub>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
</mrow>
(4) view of structural data missing is positioned, deletes the related value of the view, set to 0;Ignore for structure
Change the data of view missing, only consider that extraction does not lack the feature of view, and then complete the missing number based on multitask multi views
According to classification;
Step 2: the multitask multi views classification of structured view missing data:
It is updated to deleting the object function after lacking view correlation, selects suitable grader, train the phase of grader
Related parameter, solved using MMLE frameworks, obtain the classification results of structuring missing data;
Including following processing step:
(1) it is V={ v to define view1,v2,......vm, task is T={ t1,t2,......tn, the data point after mapping isThen the object function of complete data is:
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</munder>
<msubsup>
<mi>w</mi>
<mrow>
<mi>t</mi>
<mo>,</mo>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mi>v</mi>
</msubsup>
<mo>|</mo>
<mo>|</mo>
<msubsup>
<mi>y</mi>
<mrow>
<mi>t</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
<mi>v</mi>
</msubsup>
<mo>-</mo>
<msubsup>
<mi>y</mi>
<mrow>
<mi>t</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
<mi>v</mi>
</msubsup>
<mo>|</mo>
<msup>
<mo>|</mo>
<mn>2</mn>
</msup>
</mrow>
Wherein,WithI-th and j-th of data point under task t v-th of view are represented respectively;Represent in task
Weight in t v-th of view between data point i and j;
(2) view of definition structure missing data is vl, in the case of structural data lacks, if the data of the view are complete
Portion can not know, therefore directly leave out the value related to the view and set to 0, and renewal object function is:
<mrow>
<munder>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mrow>
<mi>v</mi>
<mo>&NotEqual;</mo>
<mi>l</mi>
</mrow>
</munder>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</munder>
<msubsup>
<mi>W</mi>
<mrow>
<mi>t</mi>
<mo>,</mo>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mi>v</mi>
</msubsup>
<mo>|</mo>
<mo>|</mo>
<msubsup>
<mi>y</mi>
<mrow>
<mi>t</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
<mi>v</mi>
</msubsup>
<mo>-</mo>
<msubsup>
<mi>y</mi>
<mrow>
<mi>t</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
<mi>v</mi>
</msubsup>
<mo>|</mo>
<msup>
<mo>|</mo>
<mn>2</mn>
</msup>
</mrow>
(3) optimal classification device is selected, such as:KNN graders are classified to sample point, obtain projection after each sample point between
Distance, it is determined that k neighbours of each data point, division classification, the distance metric of sample point are as follows:
<mrow>
<msub>
<mi>L</mi>
<mi>p</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msup>
<mrow>
<mo>(</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>l</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<mo>|</mo>
<msubsup>
<mi>y</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>l</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>-</mo>
<msubsup>
<mi>y</mi>
<mi>j</mi>
<mrow>
<mo>(</mo>
<mi>l</mi>
<mo>)</mo>
</mrow>
</msubsup>
<msup>
<mo>|</mo>
<mi>p</mi>
</msup>
<mo>)</mo>
</mrow>
<mfrac>
<mn>1</mn>
<mi>p</mi>
</mfrac>
</msup>
</mrow>
Wherein, l represents the dimension of sample data;P represents the norm of distance metric;Lp(yi,yj) represent sample yiWith yjBetween
Lp distances.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710990955.XA CN107729942A (en) | 2017-10-23 | 2017-10-23 | A kind of sorting technique of structured view missing data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710990955.XA CN107729942A (en) | 2017-10-23 | 2017-10-23 | A kind of sorting technique of structured view missing data |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107729942A true CN107729942A (en) | 2018-02-23 |
Family
ID=61212286
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710990955.XA Pending CN107729942A (en) | 2017-10-23 | 2017-10-23 | A kind of sorting technique of structured view missing data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107729942A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110378744A (en) * | 2019-07-25 | 2019-10-25 | 中国民航大学 | Civil aviaton's frequent flight passenger value category method and system towards incomplete data system |
CN110543916A (en) * | 2019-09-06 | 2019-12-06 | 天津大学 | Method and system for classifying missing multi-view data |
CN111580083A (en) * | 2020-04-30 | 2020-08-25 | 北京荣达千里科技有限公司 | Flight target threat degree identification method and system based on decision tree and storage medium |
-
2017
- 2017-10-23 CN CN201710990955.XA patent/CN107729942A/en active Pending
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110378744A (en) * | 2019-07-25 | 2019-10-25 | 中国民航大学 | Civil aviaton's frequent flight passenger value category method and system towards incomplete data system |
CN110543916A (en) * | 2019-09-06 | 2019-12-06 | 天津大学 | Method and system for classifying missing multi-view data |
CN111580083A (en) * | 2020-04-30 | 2020-08-25 | 北京荣达千里科技有限公司 | Flight target threat degree identification method and system based on decision tree and storage medium |
CN111580083B (en) * | 2020-04-30 | 2023-10-10 | 北京荣达千里科技有限公司 | Decision tree-based flying target threat degree identification method, system and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109508360B (en) | Geographical multivariate stream data space-time autocorrelation analysis method based on cellular automaton | |
CN110516539A (en) | Remote sensing image building extracting method, system, storage medium and equipment based on confrontation network | |
Bremer et al. | Interactive exploration and analysis of large-scale simulations using topology-based data segmentation | |
CN107273429B (en) | A kind of Missing Data Filling method and system based on deep learning | |
CN101093559B (en) | Method for constructing expert system based on knowledge discovery | |
CN109165664A (en) | A kind of attribute missing data collection completion and prediction technique based on generation confrontation network | |
CN113032238B (en) | Real-time root cause analysis method based on application knowledge graph | |
CN105869173A (en) | Stereoscopic vision saliency detection method | |
CN107833213A (en) | A kind of Weakly supervised object detecting method based on pseudo- true value adaptive method | |
CN103942841A (en) | Mineral resource multivariate information processing method and system based on GIS | |
CN106599230A (en) | Method and system for evaluating distributed data mining model | |
CN107729942A (en) | A kind of sorting technique of structured view missing data | |
CN108647691A (en) | A kind of image classification method based on click feature prediction | |
CN111275171A (en) | Small target detection method based on parameter sharing and multi-scale super-resolution reconstruction | |
CN109934615A (en) | Product marketing method based on depth sparse network | |
CN107730553A (en) | A kind of Weakly supervised object detecting method based on pseudo- true value search method | |
CN109614896A (en) | A method of the video content semantic understanding based on recursive convolution neural network | |
CN109741301A (en) | A kind of intensive object count method based on deep learning faster | |
CN106991049A (en) | A kind of Software Defects Predict Methods and forecasting system | |
CN106097094A (en) | A kind of man-computer cooperation credit evaluation new model towards medium-sized and small enterprises | |
CN102609721B (en) | Remote sensing image clustering method | |
CN104102716A (en) | Imbalance data predicting method based on cluster stratified sampling compensation logic regression | |
CN105761152A (en) | Topic participation prediction method based on triadic group in social network | |
CN114265954B (en) | Graph representation learning method based on position and structure information | |
Sun et al. | Automatic building age prediction from street view images |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180223 |