CN112184056A - Data feature extraction method and system based on convolutional neural network - Google Patents
Data feature extraction method and system based on convolutional neural network Download PDFInfo
- Publication number
- CN112184056A CN112184056A CN202011115965.7A CN202011115965A CN112184056A CN 112184056 A CN112184056 A CN 112184056A CN 202011115965 A CN202011115965 A CN 202011115965A CN 112184056 A CN112184056 A CN 112184056A
- Authority
- CN
- China
- Prior art keywords
- index
- data
- index set
- preprocessing
- indexes
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 54
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 49
- 238000007781 pre-processing Methods 0.000 claims abstract description 95
- 238000012545 processing Methods 0.000 claims abstract description 33
- 238000013528 artificial neural network Methods 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 16
- 238000012804 iterative process Methods 0.000 claims description 10
- 238000000034 method Methods 0.000 abstract description 13
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 238000012549 training Methods 0.000 description 13
- 238000010586 diagram Methods 0.000 description 12
- 239000000284 extract Substances 0.000 description 10
- 230000008859 change Effects 0.000 description 6
- 238000012163 sequencing technique Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 230000007547 defect Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000002596 correlated effect Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000000875 corresponding effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000007306 turnover Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- General Engineering & Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Evolutionary Computation (AREA)
- Economics (AREA)
- Artificial Intelligence (AREA)
- Educational Administration (AREA)
- Life Sciences & Earth Sciences (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Computing Systems (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a data feature extraction method and system based on a convolutional neural network, and belongs to the technical field of artificial intelligence. The method comprises the following steps: generating a preprocessing index set according to a plurality of data indexes, and executing iterative processing: taking out the head data indexes of the preprocessing index set and putting the head data indexes at the tail of the fine index set; selecting a second data index from the index set with the head data index taken out according to the head data index and then putting the second data index into the end of the fine index set; selecting a third data index from the index set after the second data index is taken out according to the head data index, the second data index and the current iteration frequency, and then putting the third data index at the end of the fine index set; determining the index set after the third data index is taken out as a preprocessing index set, and continuously executing iterative processing until the current preprocessing index set is an empty set; the fine index set and the preprocessing index set before the iterative processing are spliced into a target data set and then input into a convolutional neural network, so that the effective data characteristics of the data indexes can be efficiently and conveniently extracted.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a data feature extraction method and system based on a convolutional neural network.
Background
Convolutional neural networks are a type of deep learning network that has been gaining wide attention in recent years. The traditional convolutional neural network mainly utilizes the characteristics of local association, weight sharing and the like in adjacent regions to extract features, and compared with a fully-connected neural network, the traditional convolutional neural network can reserve important parameters to the maximum extent and filter a large number of unimportant parameters. The feature extraction method is widely used in the field of computer vision, can extract useful high-quality features as much as possible, avoids complex pre-processing of images, and has limitations when a convolutional neural network is applied in the field of ectopic correlation data (such as financial data) analysis. Ectopic correlation data is data in which the correlation between different attributes in a data set is independent of the location in its storage space. In the image recognition process, data are formed according to image pixels, and the physical position of the image data with larger dependence is extracted by the characteristics of the data; if the relative position of the data changes, the original features of the data also change, so that the image data does not belong to the ectopic correlation data. And the correlation between different indexes of the financial data is irrelevant to the position in the storage space, if the relative position of different financial data is changed on the index dimension, the correlation is still unchanged, so the financial data is ectopic correlation data. Because different data indexes are associated in different degrees, important associated features among the indexes need to be extracted in the feature extraction process, and the traditional convolution mode only carries out convolution processing on local parts, so that partial associated features are easy to lose, the global associated relationship is difficult to obtain, and the extraction of effective features is incomplete. Therefore, when a convolutional neural network is applied to the ectopic correlation data, the data association information between non-parts can be lost.
Disclosure of Invention
The embodiment of the invention mainly aims to provide a data feature extraction method and system based on a convolutional neural network, so as to efficiently and conveniently extract effective data features, and further improve the precision and efficiency of subsequent model training.
In order to achieve the above object, an embodiment of the present invention provides a data feature extraction method based on a convolutional neural network, including:
acquiring a plurality of data indexes, and generating a pretreatment index set according to the plurality of data indexes;
the following iterative process is performed:
taking out the head data indexes of the preprocessing index set and then putting the head data indexes at the tail of the fine index set;
selecting a second data index from the index set after the head data index is taken out according to the head data index, and taking out the second data index and then putting the second data index into the tail of the fine index set;
selecting a third data index from the index set after the second data index is taken out according to the head data index, the second data index and the current iteration frequency, and putting the third data index into the tail of the fine index set after the third data index is taken out;
determining the index set after the third data index is taken out as a preprocessing index set, and continuing to execute iterative processing until the current preprocessing index set is an empty set;
splicing the fine index set and the preprocessing index set before executing the iterative processing into a target data set;
and inputting the target data set into a convolution neural network obtained in advance to obtain the data characteristics of the data indexes.
The embodiment of the present invention further provides a data feature extraction system based on a convolutional neural network, including:
the preprocessing index set unit is used for acquiring a plurality of data indexes and generating a preprocessing index set according to the data indexes;
a fine index set unit for performing an iterative process of:
taking out the head data indexes of the preprocessing index set and then putting the head data indexes at the tail of the fine index set;
selecting a second data index from the index set after the head data index is taken out according to the head data index, and taking out the second data index and then putting the second data index into the tail of the fine index set;
selecting a third data index from the index set after the second data index is taken out according to the head data index, the second data index and the current iteration frequency, and putting the third data index into the tail of the fine index set after the third data index is taken out;
determining the index set after the third data index is taken out as a preprocessing index set, and continuing to execute iterative processing until the current preprocessing index set is an empty set;
the target data set unit is used for splicing the fine index set and the preprocessing index set before executing the iterative processing into a target data set;
and the data feature extraction unit is used for inputting the target data set into a convolution neural network obtained in advance to obtain the data features of the data indexes.
The embodiment of the invention also provides computer equipment which comprises a memory, a processor and a computer program which is stored on the memory and runs on the processor, wherein the processor realizes the steps of the data feature extraction method based on the convolutional neural network when executing the computer program.
The embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the data feature extraction method based on the convolutional neural network are implemented.
The data feature extraction method and system based on the convolutional neural network can efficiently and conveniently extract effective data features, avoid loss of important ectopic correlation data, and further improve precision and efficiency of subsequent model training.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of a convolutional neural network-based data feature extraction method in an embodiment of the present invention;
FIG. 2 is a flow chart of generating a fine index set in an embodiment of the present invention;
FIG. 3 is a schematic diagram of generating a fine metric set according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of convolution in an embodiment of the present invention;
FIG. 5 is a flow chart of generating a set of pre-processing metrics in an embodiment of the present invention;
FIG. 6 is a flow chart of generating a fine metrics set in another embodiment of the present invention;
FIG. 7 is a diagram illustrating the generation of a set of pre-processing metrics, in accordance with an embodiment of the present invention;
FIG. 8 is a flow chart of determining index relevance in an embodiment of the present invention;
FIG. 9 is a diagram illustrating the financial fraud identification results obtained in an embodiment of the present invention;
FIG. 10 is a block diagram of a convolutional neural network-based data feature extraction system according to an embodiment of the present invention;
fig. 11 is a block diagram of a computer device in the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.
As will be appreciated by one skilled in the art, embodiments of the present invention may be embodied as a system, apparatus, device, method, or computer program product. Accordingly, the present disclosure may be embodied in the form of: entirely hardware, entirely software (including firmware, resident software, micro-code, etc.), or a combination of hardware and software.
In view of the fact that in the prior art, only partial convolution processing is performed on local parts, partial correlation characteristics are easy to lose, a global correlation relationship is difficult to obtain, and effective characteristic extraction is incomplete, the embodiment of the invention provides a data characteristic extraction method based on a convolutional neural network, so that effective data characteristics can be extracted efficiently and conveniently, and the precision and efficiency of subsequent model training are improved. The present invention will be described in detail below with reference to the accompanying drawings.
In order to solve the problem that the traditional convolutional network is applied to the defect that local features are excessively concerned in the process of extracting the ectopic correlation data features, the invention provides a data feature extraction method based on a non-local packet convolutional neural network to improve the quality of data feature extraction, avoid losing important correlation features among different indexes due to a data physical arrangement mode, and simultaneously avoid reducing training efficiency due to data redundancy caused by excessive cross of data. The invention extracts the characteristics of the ectopic correlation data, specifically utilizes the characteristic that different data indexes have different degrees of correlation, and forms temporary local data in each combination by performing cross grouping on the data with different dimensions, thereby extracting the characteristics of the grouping by utilizing the advantage of the characteristic extraction of the local data of the convolutional network, and finally obtaining the high-quality effective data characteristics. The non-local packet cross-convolution neural network is more suitable for the field of ectopic correlation data (such as financial data) analysis, and is a more efficient and convenient feature extraction means in a specific scene.
In order to achieve the purpose, the invention is realized by the following technical scheme:
1. and sorting the required data according to a time sequence to obtain data indexes which can be analyzed in all the data.
2. And performing correlation analysis on the data indexes in the time sequence, analyzing the correlation degree among the data indexes, sequencing the indexes with lower correlation degree in sequence, and taking the sequenced data indexes as a preprocessed index data set.
3. And taking the preprocessed index data set as a data set 1, then performing cross grouping, and generating a fine index set (a data set 2) according to a sorting algorithm.
4. And splicing the preprocessing index set (data group 1) and the fine index set (data group 2) into a target data set, and then performing convolution operation to obtain a final feature extraction result.
Fig. 1 is a flowchart of a convolutional neural network-based data feature extraction method according to an embodiment of the present invention. FIG. 2 is a flow chart of generating a fine metrics set in an embodiment of the present invention. FIG. 3 is a schematic diagram of generating a fine metrics set in an embodiment of the present invention. As shown in fig. 1 to 3, the data feature extraction method based on the convolutional neural network includes:
s101: and acquiring a plurality of data indexes, and generating a pretreatment index set according to the plurality of data indexes.
Considering that only local correlation is performed on an array sequence in the conventional convolution operation, the extracted features have physical continuity, but the correlation between different data indexes needs to be extracted from the ectopic correlation data, and in order to make up for the defect, a preprocessing index set (data set 1) needs to be finely grouped. In order to preserve as much important parameters as possible, with reference to the physical position in the conventional convolution operation, continuity in the physical position should be avoided as much as possible at the time of fine grouping. Grouping procedure As shown in FIGS. 2 and 3, data group 1(G1) includes G1[ G1 ]10,g11,g12,...,g1n]Here, three pointers can be used to indicate the location of each pointer packet, using the pointer head i (head data pointer), the pointer tail j (second data pointer), and the pointer middle ═ i + j/2 + [ (-1) ^ (h-2)]X (h-1) (third data index), h is the number of iterations (number of cycles). In the sorting process, every 3 data indexes are taken as one iteration (loop).
For example, the following iterative process may be performed:
s102: and taking out the head data indexes of the preprocessing index set and then putting the head data indexes at the tail of the fine index set.
In the first iteration, G1[ G ] is selected10]Header data index G2[ G ] as a fine index set (data group 2)20]I.e. the pointer header i. The initial set of fine indicators is empty, hence in the first iterationThe head data index is a head data index of the fine index set. The fine index set in the second iteration includes three data indexes arranged in the put order, so the head data index (the first data index in the pre-processing index set) in the second iteration is the fourth data index G2[ G ] in the fine index set23]。
S103: and selecting a second data index from the index set after the head data index is taken out according to the head data index, and taking out the second data index and then putting the second data index into the tail of the fine index set.
In an embodiment, selecting the second data index from the index set after the head data index is extracted according to the head data index includes: determining an index in the preprocessing index set which is farthest from the head data index as a second data index; and selecting a second data index from the index set after the head data index is taken out.
For example, in the first iteration, G1[ G ] is chosen10]G1[ G ] farthest in the physical position to which its data group G1 corresponds1n]As G2[ G21]I.e. the pointer tail j. When the second data indicator is placed at the end of the fine indicator set, the fine indicator set is populated with the header data indicator G2[ G ] of the set of pre-processing indicators20]Placing the second data indicator at the end of the fine indicator set is equivalent to placing the second data indicator at the header data indicator G2[ G20]Behind the head.
S104: and selecting a third data index from the index set after the second data index is taken out according to the head data index, the second data index and the current iteration frequency, and taking out the third data index and then placing the third data index at the tail of the fine index set.
For example, the position G2[ G ] is chosen in the first iteration22]Selects the two data indexes (G2[ G ]) simultaneously20]And G2[ G21]) Corresponding to a physical location (G1[ G ] G) in data set 1 (set of preprocessing indices) G110]And G1[ G1n]) The farthest position, i.e. middle pointer ═ (i + j)/2+ [ (-1) ^ (h-2)]X (h-1), h is the number of iterations. Wherein, when i + j is odd, (i + j)/2 is an integer downwards.
In each iteration, the data indexes (the head data index, the second data index and the third data index) are put at the end of the fine index set, which is equivalent to arranging the data indexes in sequence according to the order of putting the data indexes into the fine index set.
S105: and determining the index set after the third data index is taken out as a preprocessing index set.
S106: and judging whether the current preprocessing index set is an empty set.
S107: and when the current preprocessing index set is an empty set, splicing the fine index set and the preprocessing index set before executing iterative processing into a target data set.
And when the current preprocessing index set is not an empty set, returning to the step S102.
As shown in fig. 2, when the current set of pre-processing indicators is an empty set, it indicates that the number of data indicators in the fine indicator set G2 (Len (G2)) is equal to the number of data indicators in the pre-processing indicator set G1 before performing iteration processing (Len (G1)), the length of the fine indicator set G2 is equal to the length of the pre-processing indicator set G1, the iteration is ended, the sorting is stopped, and at this time, the final fine indicator set G2 is obtained.
When the current set of pre-processing indexes is not an empty set, it indicates that the number of data indexes in the fine index set G2 (Len (G2)) is smaller than the number of data indexes in the pre-processing index set G1 before the iterative processing is performed (Len (G1)), and the length of the fine index set G2 is smaller than that of the pre-processing index set G1.
Wherein S105 to S107 correspond to: and determining the index set after the third data index is taken out as a preprocessing index set, continuously executing iterative processing until the current preprocessing index set is an empty set, and splicing the fine index set and the preprocessing index set before executing the iterative processing into a target data set.
Fig. 4 is a schematic diagram of convolution in an embodiment of the present invention. As shown in part a of fig. 4, G0 is raw data, G1 is a set of pre-processed metrics (data set 1), and G2 is a set of fine metrics G2 (data set 2). G1 and G2 are spliced to obtain a data group G3[ G10,g11,...,g1n,g20,g21,...,g2n]。
S108: and inputting the target data set into a convolution neural network obtained in advance to obtain the data characteristics of the data indexes.
As shown in part b of fig. 4, G3 is used as Input to the convolution operation step (Input), and the convolution network part is composed of 3 convolution layers. The convolution layer uses a small convolution kernel for continuous convolution operation, the size of the convolution kernel is 3 Xm, 3 is an index dimension size, m is a time dimension size (in fig. 4, a horizontal axis X of a coordinate system is an index, and a vertical axis T is time), m can be flexibly adjusted and set according to a specific scene, and the fixed step length of convolution is 1. To prevent the index feature from being lost, 1 data (0 is filled) is filled in the edge of the index dimension data, and no filling operation is performed on the time dimension data. And outputting to obtain a final data characteristic (Output) after the convolution operation.
The main body of the convolutional neural network-based data feature extraction method shown in fig. 1 may be a computer. As can be seen from the flow shown in fig. 1, the data feature extraction method based on the convolutional neural network according to the embodiment of the present invention can efficiently and conveniently extract effective data features, avoid loss of important ectopic correlation data, and further improve the accuracy and efficiency of subsequent model training.
FIG. 5 is a flow chart of generating a set of pre-processing metrics in an embodiment of the present invention. FIG. 6 is a flow chart of generating a fine index set in another embodiment of the present invention. FIG. 7 is a diagram illustrating the generation of a set of pre-processing metrics, according to an embodiment of the present invention. As shown in fig. 5-7, generating a set of pre-processing metrics from a plurality of data metrics includes:
s201: an original index set is generated from the plurality of data indices.
In specific implementation, the data are sorted into time series according to time units (such as month, quarter, etc.), and data indexes capable of being analyzed are selected from all the data. For example, the data indicators may include revenue, accounts receivable, inventory, asset profitability, and asset liability rate, among others.
S202: randomly taking one of the data indexes from the original index set, and determining the data index as a reference index.
For example, the original index set G0 has n +1 indexes, and one index is randomly selected as a reference index (b)g10)。
S203: the following iterative process is performed: putting the reference index at the tail of the pretreatment index set; and determining the data index with the lowest correlation degree with the reference index in the original index set after the reference index is taken out as the reference index, and continuing to execute iterative processing until the original index set is an empty set.
And each iteration puts the reference indexes at the tail of the preprocessing index set, which is equivalent to arranging the data indexes in sequence according to the sequence of putting the data indexes into the preprocessing index set. As shown in fig. 6-7, in a specific implementation, a pointer is set to the reference index, and the reference index is used as a header (index is 10, index is a subscript of data group 1) of the set of pre-processing indexes (data group 1); the initial set of pre-processing indicators is an empty set, which is equivalent to putting the reference indicators at the end of the set of pre-processing indicators. The subsequent indexes are sorted according to the correlation among the indexes. Selecting the data index with the lowest relevance degree with the reference index as the next data index (index is 11) of the preprocessing index set (data group 1), and moving the pointer to the tail part once to move the next data index (g)11Index 11) as a reference index, and similarly, an index having the lowest correlation with the reference index (index 11) is selected as a next data index (g)12Index 12), wherein the sorted data indexes that have been completed do not participate in the correlation calculation (which is equivalent to determining the data index with the lowest correlation with the reference index in the original index set after the reference index is taken out as the reference index), and repeating the above steps until the pointer points to the data index (g)1n Index 1n), the original index set is a null set, and finally the arrangement is completed to obtain a pre-processing index set G1 (data group 1).
FIG. 8 is a flowchart of determining index association in an embodiment of the invention. As shown in fig. 8, the data feature extraction method based on the convolutional neural network further includes:
s301: and determining the covariance between the reference index and each data index in the original index set after the reference index is taken out.
In specific implementation, if the reference index is U and another index in the original index set is V, the covariance between U and V is as follows:
Cov(U,V)＝E{[U-E(U)][V-E(V)]}；
wherein Cov (U, V) is the covariance between U and V, describing the degree of correlation between two sets of variables U and V, e (U) is the expectation of U, e (V) is the expectation of V. When Cov (U, V) is greater than 0, U is positively correlated with V; when Cov (U, V) is less than 0, U is negatively correlated with V; when Cov (U, V) is equal to 0, U is not correlated with V.
S302: the variance of the reference index and the variance of each data index are determined.
S303: and determining the association degree of each data index and the reference index in the original index set after the reference index is taken out according to the covariance between the reference index and each data index, the variance of the reference index and the variance of each data index.
For example, the degree of association (Pearson correlation coefficient) between the reference index U and another index V in the original index set is as follows:
where ρ (U, V) is the degree of association between U and V, Var (U) is the variance of U, and Var (V) is the variance of V.
When rho (U, V) is 1, U and V are in complete positive correlation, and V is aU + b, wherein a is more than 0;
when rho (U, V) is-1, U and V are completely negative correlation, and satisfy the condition that V is-aU + b, and a is more than 0;
when 0 < | ρ (U, V) | < 1, U has some correlation with V.
The specific process of the embodiment of the invention is as follows:
1. and acquiring a plurality of data indexes, and generating an original index set according to the plurality of data indexes.
2. Randomly taking one of the data indexes from the original index set, and determining the data index as a reference index.
3. And putting the reference index at the tail of the preprocessing index set.
4. And determining the covariance between the reference index and each data index in the original index set after the reference index is taken out, and determining the variance of the reference index and the variance of each data index.
5. And determining the association degree of each data index and the reference index in the original index set after the reference index is taken out according to the covariance between the reference index and each data index, the variance of the reference index and the variance of each data index.
6. Judging whether the original index set is an empty set or not; and (4) executing step 7 if the original index set is an empty set, otherwise, determining the data index with the lowest correlation degree with the reference index in the original index set after the reference index is taken out as the reference index, and returning to execute step 3.
The following iterative process is performed:
7. taking out the head data indexes of the preprocessing index set and then putting the head data indexes at the tail of the fine index set;
8. and determining an index which is farthest away from the head data index in the preprocessing index set as a second data index, and selecting the second data index from the index set after the head data index is taken out.
9. And selecting a third data index from the index set after the second data index is taken out according to the head data index, the second data index and the current iteration frequency, and taking out the third data index and then placing the third data index at the tail of the fine index set.
10. And determining the index set after the third data index is taken out as a preprocessing index set.
11. And judging whether the current preprocessing index set is an empty set. And when the current preprocessing index set is an empty set, splicing the preprocessing index set and the fine index set before the iterative processing is executed into a target data set. And when the current preprocessing index set is not an empty set, returning to the step 7.
12. And inputting the target data set into a convolution neural network obtained in advance to obtain the data characteristics of the data indexes.
FIG. 9 is a diagram illustrating the financial fraud identification results obtained in an embodiment of the present invention. As shown in fig. 9, when identifying whether a financial affair is counterfeit, a financial index with a high correlation with financial counterfeit is selected from the financial data, and the finally selected 11 data indexes with significant differences are: the method comprises the steps of carrying out inventory balance adjustment on the basis of the original index set G0, wherein the original index set G0 comprises 11 groups of time sequences, and is characterized by comprising the following steps of asset turnover rate, asset earning rate, asset liability rate, accounts receivable ratio, accounts receivable change rate, other accounts receivable ratio, other accounts receivable change rate, inventory ratio, stock change rate, cash sale rate and free cash flow change.
Sequencing G0 by determining the relevance of each data index in G0 to obtain a preprocessing index set G1, wherein the purpose is to reorder G0, so that each adjacent index in G1 has lower linear relevance with each other, and thus, the nonlinear relevance characteristics are identified through a subsequent convolution neural network; and sequencing G1 in a grouping sequencing mode to obtain a fine index set G2, wherein the purpose is to reorder G1, aggregate indexes which are far away from each other in physical position on a storage space, form temporary local data and further extract nonlinear correlation. G2 can be said to be a supplement of G1, and can make up for the defect that G1 has insufficient extraction of correlation among partial indexes. The data contents stored by the three groups of data of G0, G1 and G2 are the same, and are all the 11 index sequences, and the three groups of differences are reflected in that the physical positions of different indexes on the storage space are different in index dimension.
G1 and G2 are spliced and input into a convolutional neural network with a convolution kernel of 3 xn and a step length of 1, wherein 3 is the direction of an index dimension (X), n is the direction of a time dimension (T), and n is a variable parameter, and is determined according to actual conditions. This step corresponds to part b of fig. 9.
The final Output part of the convolutional neural network is the extracted target Features (Output Features) of the present invention, and the Output Features usually need to be input into a subsequent specific scene model. Taking the financial counterfeiting recognition model as an example, the data features obtained after the processing of the invention can be used as the input of the LSTM neural network, the obtained features are input into the corresponding LSTM neural network according to the time sequence direction, and the final financial counterfeiting recognition result is obtained through the learning training of the network. The part corresponds to part c of fig. 9, wherein LSTM is a long-short memory neural network, FC is a fully-connected neural network, and the financial counterfeiting identification model comprises a convolutional neural network part b and a differential model part c in the graph. In the identification process, the data index set to be identified is input to the part a as G0, after sequencing and grouping are carried out, the trained part b and part c are input, and finally, the financial counterfeiting identification result (Output Results) is Output. Compared with the prior art that the original data set G0 is directly input to the part b, the method can more effectively extract the incidence relation among different indexes, and enables the common convolution operation to generate better effect.
In conclusion, the data set 1 of the invention realizes the initial sequencing of the original data indexes, and forms the data indexes with low linear correlation degree into local combination so as to realize the optimization of the correlation characteristics of the initial local information; the data set 2 realizes the change of the data set 1 on the physical position, and the association between the local data indexes is recombined and corresponds to the association between the data set 1 and the non-local data indexes. Each local data index in the data set 2 is recombined by the non-local data indexes in the data set 1, so that each convolution kernel can convolve the non-local area of the layer above the local data index, thereby realizing the non-local feature extraction at a longer distance on the physical position. Meanwhile, the invention only generates a group of optimal packets, thereby avoiding the efficiency from being greatly reduced due to excessive packets. After the convolution operation is finished, the data characteristics are finally obtained, the correlation characteristics among the data indexes are retained to a greater extent, and the training efficiency is ensured to a certain extent. The method is the basic steps of various subsequent model training, and is also a key factor influencing the model training precision.
The invention improves the characteristic extraction effect through an index grouping algorithm, so that the grouped data has the advantage of being more suitable for local characteristic identification of the convolutional neural network, effectively avoids the defect that the convolutional neural network excessively pays attention to local characteristics when extracting the index characteristics, simultaneously avoids the reduction of training efficiency and extraction efficiency caused by data redundancy due to excessive grouping of the data, and provides a better characteristic extraction scheme for scenes needing deep analysis on (financial) data.
Based on the same inventive concept, the embodiment of the invention also provides a data feature extraction system based on the convolutional neural network, and as the problem solving principle of the system is similar to the data feature extraction method based on the convolutional neural network, the implementation of the system can refer to the implementation of the method, and repeated parts are not repeated.
Fig. 10 is a block diagram of a data feature extraction system based on a convolutional neural network in the embodiment of the present invention. As shown in fig. 10, the convolutional neural network-based data feature extraction system includes:
the preprocessing index set unit is used for acquiring a plurality of data indexes and generating a preprocessing index set according to the data indexes;
a fine index set unit for performing an iterative process of:
taking out the head data indexes of the preprocessing index set and then putting the head data indexes at the tail of the fine index set;
selecting a second data index from the index set after the head data index is taken out according to the head data index, and taking out the second data index and then putting the second data index into the tail of the fine index set;
selecting a third data index from the index set after the second data index is taken out according to the head data index, the second data index and the current iteration frequency, and putting the third data index into the tail of the fine index set after the third data index is taken out;
determining the index set after the third data index is taken out as a preprocessing index set, and continuing to execute iterative processing until the current preprocessing index set is an empty set;
the target data set unit is used for splicing the fine index set and the preprocessing index set before executing the iterative processing into a target data set;
and the data feature extraction unit is used for inputting the target data set into a convolution neural network obtained in advance to obtain the data features of the data indexes.
In one embodiment, the preprocessing index set unit is specifically configured to:
generating an original index set according to a plurality of data indexes;
randomly taking one data index from the original index set, determining the data index as a reference index, and executing the following iterative processing:
putting the reference index at the tail of the pretreatment index set;
and determining the data index with the lowest correlation degree with the reference index in the original index set after the reference index is taken out as the reference index, and continuously executing iterative processing until the original index set is an empty set.
In one embodiment, the fine metric set unit is specifically configured to:
determining an index in the preprocessing index set which is farthest from the head data index as a second data index;
and selecting a second data index from the index set after the head data index is taken out.
In one embodiment, the preprocessing index set unit is further configured to:
determining the covariance between the reference index and each data index in the original index set after the reference index is taken out;
determining the variance of the reference index and the variance of each data index;
and determining the association degree of each data index and the reference index in the original index set after the reference index is taken out according to the covariance between the reference index and each data index, the variance of the reference index and the variance of each data index.
In conclusion, the data feature extraction system based on the convolutional neural network can efficiently and conveniently extract effective data features, avoid loss of important ectopic correlation data, and further improve precision and efficiency of subsequent model training.
The embodiment of the invention also provides a specific implementation mode of computer equipment capable of realizing all the steps in the data feature extraction method based on the convolutional neural network in the embodiment. Fig. 11 is a block diagram of a computer device in an embodiment of the present invention, and referring to fig. 11, the computer device specifically includes the following contents:
a processor (processor)1101 and a memory (memory) 1102.
The processor 1101 is configured to call a computer program in the memory 1102, and the processor implements all the steps in the data feature extraction method based on the convolutional neural network in the above embodiment when executing the computer program, for example, the processor implements the following steps when executing the computer program:
acquiring a plurality of data indexes, and generating a pretreatment index set according to the plurality of data indexes;
the following iterative process is performed:
taking out the head data indexes of the preprocessing index set and then putting the head data indexes at the tail of the fine index set;
selecting a second data index from the index set after the head data index is taken out according to the head data index, and taking out the second data index and then putting the second data index into the tail of the fine index set;
selecting a third data index from the index set after the second data index is taken out according to the head data index, the second data index and the current iteration frequency, and putting the third data index into the tail of the fine index set after the third data index is taken out;
determining the index set after the third data index is taken out as a preprocessing index set, and continuing to execute iterative processing until the current preprocessing index set is an empty set;
splicing the fine index set and the preprocessing index set before executing the iterative processing into a target data set;
and inputting the target data set into a convolution neural network obtained in advance to obtain the data characteristics of the data indexes.
In conclusion, the computer equipment provided by the embodiment of the invention can efficiently and conveniently extract effective data characteristics, avoid the loss of important ectopic correlation data, and further improve the precision and efficiency of subsequent model training.
An embodiment of the present invention further provides a computer-readable storage medium capable of implementing all the steps in the data feature extraction method based on the convolutional neural network in the above embodiment, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements all the steps of the data feature extraction method based on the convolutional neural network in the above embodiment, for example, when the processor executes the computer program, the processor implements the following steps:
acquiring a plurality of data indexes, and generating a pretreatment index set according to the plurality of data indexes;
the following iterative process is performed:
taking out the head data indexes of the preprocessing index set and then putting the head data indexes at the tail of the fine index set;
selecting a second data index from the index set after the head data index is taken out according to the head data index, and taking out the second data index and then putting the second data index into the tail of the fine index set;
selecting a third data index from the index set after the second data index is taken out according to the head data index, the second data index and the current iteration frequency, and putting the third data index into the tail of the fine index set after the third data index is taken out;
determining the index set after the third data index is taken out as a preprocessing index set, and continuing to execute iterative processing until the current preprocessing index set is an empty set;
splicing the fine index set and the preprocessing index set before executing the iterative processing into a target data set;
and inputting the target data set into a convolution neural network obtained in advance to obtain the data characteristics of the data indexes.
In summary, the computer-readable storage medium of the embodiment of the present invention can efficiently and conveniently extract effective data features, avoid loss of important ectopic correlation data, and further improve the precision and efficiency of subsequent model training.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Those of skill in the art will further appreciate that the various illustrative logical blocks, units, and steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate the interchangeability of hardware and software, various illustrative components, elements, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design requirements of the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the embodiments of the present invention.
The various illustrative logical blocks, or elements, or devices described in connection with the embodiments disclosed herein may be implemented or performed with a general purpose processor, a digital signal processor, an Application Specific Integrated Circuit (ASIC), a field programmable gate array or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general-purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a digital signal processor and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a digital signal processor core, or any other similar configuration.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may be stored in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. For example, a storage medium may be coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC, which may be located in a user terminal. In the alternative, the processor and the storage medium may reside in different components in a user terminal.
In one or more exemplary designs, the functions described above in connection with the embodiments of the invention may be implemented in hardware, software, firmware, or any combination of the three. If implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media that facilitate transfer of a computer program from one place to another. Storage media may be any available media that can be accessed by a general purpose or special purpose computer. For example, such computer-readable media can include, but is not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to carry or store program code in the form of instructions or data structures and which can be read by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor. Additionally, any connection is properly termed a computer-readable medium, and, thus, is included if the software is transmitted from a website, server, or other remote source via a coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL), or wirelessly, e.g., infrared, radio, and microwave. Such discs (disk) and disks (disc) include compact disks, laser disks, optical disks, DVDs, floppy disks and blu-ray disks, where disks usually reproduce data magnetically, while disks usually reproduce data optically with lasers. Combinations of the above may also be included in the computer-readable medium.
Claims (10)
1. A data feature extraction method based on a convolutional neural network is characterized by comprising the following steps:
acquiring a plurality of data indexes, and generating a pretreatment index set according to the data indexes;
the following iterative process is performed:
taking out the head data indexes of the preprocessing index set and then putting the head data indexes at the tail of the fine index set;
selecting a second data index from the index set with the head data index taken out according to the head data index, taking out the second data index and putting the second data index at the tail of the fine index set;
selecting a third data index from the index set after the second data index is taken out according to the head data index, the second data index and the current iteration frequency, taking out the third data index and then placing the third data index at the tail of the fine index set;
determining the index set after the third data index is taken out as a preprocessing index set, and continuously executing the iterative processing until the current preprocessing index set is an empty set;
splicing the fine index set and the preprocessing index set before executing the iterative processing into a target data set;
and inputting the target data set into a convolution neural network obtained in advance to obtain the data characteristics of the data indexes.
2. The convolutional neural network-based data feature extraction method of claim 1, wherein generating a set of pre-processing metrics from the plurality of data metrics comprises:
generating an original index set according to the plurality of data indexes;
randomly taking one data index out of the original index set, determining the data index as a reference index, and executing the following iterative processing:
placing the reference index at the end of the set of pre-processing indices;
and determining the data index with the lowest correlation degree with the reference index in the original index set after the reference index is taken out as the reference index, and continuously executing the iterative processing until the original index set is an empty set.
3. The convolutional neural network-based data feature extraction method of claim 1, wherein selecting a second data index from the index set from which the head data index is extracted according to the head data index comprises:
determining, as a second data indicator, an indicator in the set of pre-processing indicators that is furthest from the head data indicator;
and selecting a second data index from the index set after the head data index is taken out.
4. The convolutional neural network-based data feature extraction method of claim 2, further comprising:
determining the covariance between the reference index and each data index in the original index set after the reference index is taken out;
determining the variance of the reference index and the variance of each data index;
and determining the association degree of each data index and the reference index in the original index set after the reference index is taken out according to the covariance between the reference index and each data index, the variance of the reference index and the variance of each data index.
5. A convolutional neural network-based data feature extraction system, comprising:
the device comprises a preprocessing index set unit, a preprocessing index set unit and a preprocessing index set unit, wherein the preprocessing index set unit is used for acquiring a plurality of data indexes and generating a preprocessing index set according to the data indexes;
a fine index set unit for performing an iterative process of:
taking out the head data indexes of the preprocessing index set and then putting the head data indexes at the tail of the fine index set;
selecting a second data index from the index set with the head data index taken out according to the head data index, taking out the second data index and putting the second data index at the tail of the fine index set;
selecting a third data index from the index set after the second data index is taken out according to the head data index, the second data index and the current iteration frequency, taking out the third data index and then placing the third data index at the tail of the fine index set;
determining the index set after the third data index is taken out as a preprocessing index set, and continuously executing the iterative processing until the current preprocessing index set is an empty set;
the target data set unit is used for splicing the fine index set and the preprocessing index set before executing the iterative processing into a target data set;
and the data characteristic extraction unit is used for inputting the target data set into a convolution neural network obtained in advance to obtain the data characteristics of the data indexes.
6. The convolutional neural network-based data feature extraction system of claim 5, wherein the preprocessing index set unit is specifically configured to:
generating an original index set according to the plurality of data indexes;
randomly taking one data index out of the original index set, determining the data index as a reference index, and executing the following iterative processing:
placing the reference index at the end of the set of pre-processing indices;
and determining the data index with the lowest correlation degree with the reference index in the original index set after the reference index is taken out as the reference index, and continuously executing the iterative processing until the original index set is an empty set.
7. The convolutional neural network-based data feature extraction system of claim 5, wherein the fine metric set unit is specifically configured to:
determining, as a second data indicator, an indicator in the set of pre-processing indicators that is furthest from the head data indicator;
and selecting a second data index from the index set after the head data index is taken out.
8. The convolutional neural network-based data feature extraction system of claim 6, wherein the preprocessing index set unit is further configured to:
determining the covariance between the reference index and each data index in the original index set after the reference index is taken out;
determining the variance of the reference index and the variance of each data index;
and determining the association degree of each data index and the reference index in the original index set after the reference index is taken out according to the covariance between the reference index and each data index, the variance of the reference index and the variance of each data index.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and running on the processor, wherein the processor when executing the computer program implements the steps of the convolutional neural network-based data feature extraction method of any one of claims 1 to 4.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the convolutional neural network-based data feature extraction method of any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011115965.7A CN112184056B (en) | 2020-10-19 | 2020-10-19 | Data feature extraction method and system based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011115965.7A CN112184056B (en) | 2020-10-19 | 2020-10-19 | Data feature extraction method and system based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112184056A true CN112184056A (en) | 2021-01-05 |
| CN112184056B CN112184056B (en) | 2024-02-09 |
Family
ID=73950777
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011115965.7A Active CN112184056B (en) | 2020-10-19 | 2020-10-19 | Data feature extraction method and system based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112184056B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108197809A (en) * | 2017-12-28 | 2018-06-22 | 浙江工业大学 | Weights based on dimension optimum translation share the real-time scheduling method of depth network |
| CN109299185A (en) * | 2018-10-18 | 2019-02-01 | 上海船舶工艺研究所(中国船舶工业集团公司第十研究所) | A kind of convolutional neural networks for timing flow data extract the analysis method of feature |
| US20190318806A1 (en) * | 2018-04-12 | 2019-10-17 | Illumina, Inc. | Variant Classifier Based on Deep Neural Networks |
| CN110619393A (en) * | 2019-09-27 | 2019-12-27 | 上海交通大学 | Traffic simulation software parameter calibration method, system and medium based on learning algorithm |
| CN111292007A (en) * | 2020-02-28 | 2020-06-16 | 中国工商银行股份有限公司 | Supplier financial risk prediction method and device |
| CN111310676A (en) * | 2020-02-21 | 2020-06-19 | 重庆邮电大学 | Video motion recognition method based on CNN-LSTM and attention |
| CN111582395A (en) * | 2020-05-13 | 2020-08-25 | 金陵科技学院 | Product quality classification system based on convolutional neural network |
-
2020
- 2020-10-19 CN CN202011115965.7A patent/CN112184056B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108197809A (en) * | 2017-12-28 | 2018-06-22 | 浙江工业大学 | Weights based on dimension optimum translation share the real-time scheduling method of depth network |
| US20190318806A1 (en) * | 2018-04-12 | 2019-10-17 | Illumina, Inc. | Variant Classifier Based on Deep Neural Networks |
| CN109299185A (en) * | 2018-10-18 | 2019-02-01 | 上海船舶工艺研究所(中国船舶工业集团公司第十研究所) | A kind of convolutional neural networks for timing flow data extract the analysis method of feature |
| CN110619393A (en) * | 2019-09-27 | 2019-12-27 | 上海交通大学 | Traffic simulation software parameter calibration method, system and medium based on learning algorithm |
| CN111310676A (en) * | 2020-02-21 | 2020-06-19 | 重庆邮电大学 | Video motion recognition method based on CNN-LSTM and attention |
| CN111292007A (en) * | 2020-02-28 | 2020-06-16 | 中国工商银行股份有限公司 | Supplier financial risk prediction method and device |
| CN111582395A (en) * | 2020-05-13 | 2020-08-25 | 金陵科技学院 | Product quality classification system based on convolutional neural network |
Non-Patent Citations (4)
| Title |
|---|
| 张景;: "基于业务流程的财务数据自动化分类系统设计", 自动化与仪器仪表, no. 01 * |
| 肖毅;熊凯伦;张希;: "基于TEI@I方法论的企业财务风险预警模型研究", 管理评论, no. 07 * |
| 蒲晓辉;: "公司信用风险评估新模型:基于Isomap的SVM模型", 财会月刊, no. 18 * |
| 裴曼如;张立文;: "基于投资者情绪和深度学习的股票价格趋势预测研究", 金融发展, no. 01 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112184056B (en) | 2024-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107545889A (en) | Suitable for the optimization method, device and terminal device of the model of pattern-recognition | |
| CN109949219B (en) | Reconstruction method, device and equipment of super-resolution image | |
| CN109829959B (en) | Facial analysis-based expression editing method and device | |
| CN112819604A (en) | Personal credit evaluation method and system based on fusion neural network feature mining | |
| CN103942571B (en) | Graphic image sorting method based on genetic programming algorithm | |
| CN109934181A (en) | Text recognition method, device, equipment and computer-readable medium | |
| CN111784699B (en) | Method and device for carrying out target segmentation on three-dimensional point cloud data and terminal equipment | |
| CN108573399A (en) | Method and its system are recommended by trade company based on transition probability network | |
| CN110490203A (en) | Image partition method and device, electronic equipment and computer readable storage medium | |
| CN111507183A (en) | Crowd counting method based on multi-scale density map fusion cavity convolution | |
| CN110428139A (en) | The information forecasting method and device propagated based on label | |
| CN111639230B (en) | Similar video screening method, device, equipment and storage medium | |
| CN110033089A (en) | Deep neural network parameter optimization method and system based on Distributed fusion algorithm | |
| CN115358809A (en) | Multi-intention recommendation method and device based on graph comparison learning | |
| CN112836602A (en) | Behavior recognition method, device, equipment and medium based on space-time feature fusion | |
| Li et al. | D2c-sr: A divergence to convergence approach for real-world image super-resolution | |
| CN113299298B (en) | Residual error unit, network and target identification method, system, device and medium | |
| CN113240259B (en) | Rule policy group generation method and system and electronic equipment | |
| WO2024060839A1 (en) | Object operation method and apparatus, computer device, and computer storage medium | |
| CN112184056B (en) | Data feature extraction method and system based on convolutional neural network | |
| CN114697127B (en) | Service session risk processing method based on cloud computing and server | |
| Lu et al. | Siamese graph attention networks for robust visual object tracking | |
| CN115564812A (en) | Stereo matching method, system, equipment and medium based on high-precision parallax refinement | |
| CN115329116A (en) | Image retrieval method based on multi-layer feature fusion | |
| CN115293872A (en) | Method for establishing risk identification model and corresponding device |
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 |