CN116168847B - Infectious disease prediction method based on optimized next generation reserve pool calculation - Google Patents
Infectious disease prediction method based on optimized next generation reserve pool calculation Download PDFInfo
- Publication number
- CN116168847B CN116168847B CN202310460666.4A CN202310460666A CN116168847B CN 116168847 B CN116168847 B CN 116168847B CN 202310460666 A CN202310460666 A CN 202310460666A CN 116168847 B CN116168847 B CN 116168847B
- Authority
- CN
- China
- Prior art keywords
- data
- time
- input data
- prediction
- infectious disease
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/80—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for detecting, monitoring or modelling epidemics or pandemics, e.g. flu
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Health & Medical Sciences (AREA)
- Public Health (AREA)
- Engineering & Computer Science (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention refers to an infectious disease prediction method based on optimized next generation reserve pool calculation, which optimizes the original NG-RC frame in a targeted structure, improves the operation efficiency under the condition of large data input, firstly divides data into two parts of a training group and a prediction group in a data processing step, has time difference under the same time scale, and obtains output weight parameters between input data and output data by utilizing an algorithm after the data is trained; under continuous input of input data, target data to be predicted is obtained by using the output weight. The invention can achieve the aim of really predicting the future infection trend and improve the overall operation efficiency of the system.
Description
Technical Field
The invention belongs to the technical field of machine learning algorithms and infectious disease prediction, and particularly relates to an infectious disease prediction method based on optimized next-generation reserve pool calculation.
Background
In the aspect of infectious disease prediction, the traditional prediction method uses algorithms such as long-term memory network, support vector machine, random forest and the like, and has the defects of low efficiency, short prediction time and the like in the infectious disease transmission prediction due to the dependence on structural parameters and data conditions. In recent years, machine learning algorithm-reserve pool calculation is excellent in nonlinear dynamics system prediction, and has the characteristics of low training cost, high efficiency, high speed and the like. However, the pool-based calculation of infectious disease prediction techniques are not yet mature enough.
Disclosure of Invention
The technical problems to be solved are as follows: the invention discloses an infectious disease prediction method based on next generation reserve pool calculation, which can get rid of the disadvantage that the traditional prediction method is seriously dependent on a plurality of structural parameters and inherent components of a model, overcomes the defect of insufficient data at a time point, and simultaneously obtains the infection trend of a target position by utilizing the infection trend of other areas so as to achieve the aim of truly predicting the future infection trend, thereby being beneficial to the related mechanism to timely master the development of epidemic situation and protecting the life health of people.
The technical scheme is as follows:
an infectious disease prediction method based on optimized next generation pool calculation, the infectious disease prediction method comprising the steps of:
s1, carrying out internal connection optimization on the basis of an original NG-RC framework structure to obtain an optimized NG-RC framework; specifically, the input dimension is set to be N, the output dimension is set to be M, and the input data is recorded as I N The output data is denoted as I M The method comprises the steps of carrying out a first treatment on the surface of the At the input data I N Randomly selecting data with the same dimension as the output data, and marking the data as I M1 The method comprises the steps of carrying out a first treatment on the surface of the At unit time t, I M1 (t) and the previous time unit I M1 (t-1) obtaining a full connection group by adopting a way of sequential multiplication; input data I for other dimensions N-M (t)，I M1 (t) and I N-M (t) and previous time Unit I N-M (t-1) multiplying in sequence to obtain a sparse connection set;
s2, collecting infectious disease data of a plurality of areas within a period of time, generating all area data sets taking time as sampling labels, dividing the data sets into two groups, wherein most area data sets are taken as input data I N A few regional data sets as target prediction output data I M The method comprises the steps of carrying out a first treatment on the surface of the Input data I N The method comprises the steps of dividing the data into two parts in the time dimension, taking the data of the early most time sequences as training input data, and taking the rest data as test input data; taking a preset time difference t in the time dimension of training p As the prediction time, target prediction data I M T is the front of (1) p Discarding the data points of time, and respectively taking the rest data as target fitting data and test comparison data of a training stage;
s3, training data I N Inputting the optimized NG-RC framework to obtain a full connection group S corresponding to the training stage l And sparse connection set S 2 The method comprises the steps of carrying out a first treatment on the surface of the Will be fully connected to group S l And sparse connection set S 2 Combining as a set of nonlinear eigenvectors S inside a reservoir nl To group nonlinear eigenvectors S nl Sum constant term S c Linear term S n2 Combined into O total A state vector;
s4, using least square method to make state vector O total With target prediction data I M Performing data matching fitting to obtain output weightHeavy W out ；
S5, inputting data I N Inputting the optimized NG-RC framework to obtain state vector O total ；
S6, the state vector O total Output weight W for training out Performing dot product multiplication to obtain output predicted data I M 。
Further, in step S2, the process of grouping all the regional data groups with time as sampling tags into two groups includes the following steps:
s21, taking time T as sampling label to obtain all regional N-dimension data group I N Divided into two groups, wherein 95% of the data group is used as training data I N 5% of the data sets are taken as target prediction data I M ；
S22, inputting training data I N The first 90% on the time series is used as continuous input, and the last 10% is used as continuous input of the test set;
s23, for M groups of target prediction data, adopting a preset time difference t p As a predicted time interval.
Further, in step S3, the state vector O total The calculation formula of (t) is: o (O) total (t)＝[S c ；S l ；S nl ] T (t)；
Wherein the constant term S is represented by a constant 1 c The method comprises the steps of carrying out a first treatment on the surface of the Use I M (t) sequence and I M (t-1) sequence combinations represent a linear term S n2 The method comprises the steps of carrying out a first treatment on the surface of the Use I M (t) sequence and I M (t-1) sequence part combination relation means nonlinear feature vector group S nl Wherein S is nl The composition of (1) is a full connection group and a sparse connection group introduced in step S1.
Further, the least square formula adopted in step S4 is:
W out ＝YX T (XX T +λII) -1
wherein Y represents target prediction data I N-M X represents a state vector O formed by input data total The method comprises the steps of carrying out a first treatment on the surface of the Lambda is the bias parameter and takes the value of 1 multiplied by 10 -7 II is an identity matrix.
Further, the obtained prediction data I M The calculation formula of (2) is as follows: i M ＝W out *O total 。
Further, in step S2, all data is preprocessed, specifically, the data is smoothed by using a savgol filter.
The beneficial effects are that:
first, the infectious disease prediction method based on optimized next generation reserve pool calculation of the present invention gets rid of the disadvantages that the conventional prediction method is severely dependent on a plurality of structural parameters and inherent components of the model itself, and overcomes the disadvantage of insufficient data at a time point.
Secondly, the infectious disease prediction method based on the optimized next generation reserve pool calculation performs structural optimization on the original next generation reserve pool calculation, so that the method meets the data operation requirement better.
Thirdly, the infectious disease prediction method based on the optimized next generation reserve pool calculation achieves the aim of really predicting the future infection trend by acquiring the infection trend of the target position through the infection trend of other areas.
Drawings
FIG. 1 is a diagram of a python-based model structure and optimized next generation pool algorithm model data calculation using an embodiment of the present invention, wherein the simulation software is python, and the platform CPU is an AMD Ryzen 9 5900HS with Radeon Graphics eight-core processor.
Fig. 2 is a graph of predicted results obtained by randomly selecting data of four country regions.
Fig. 3 is a flowchart of an infectious disease prediction method based on optimized next generation pool calculation according to an embodiment of the present invention.
Detailed Description
The following examples will provide those skilled in the art with a more complete understanding of the invention, but are not intended to limit the invention in any way.
FIG. 1 is a data flow chart and algorithm block diagram of an infectious disease prediction method based on optimized next generation pool calculation in accordance with an embodiment of the present invention. Referring to fig. 1, the prediction method includes the steps of:
s1, on the algorithm structure of the original NG-RC, internal connection optimization is carried out. The nonlinear connection process of high-dimensional input data redundancy in the reservoir formation stage is reduced, and the nonlinear connection process is converted into internal full connection of partial data and sparse connection with the rest of data. The specific details are as follows: compared with NG-RC, a part of main feature dimensions are randomly selected, and the dimension I which is the same as the target data is randomly selected in the dimensions of input data M The data are multiplied by the previous time unit in turn to obtain a part of full connection group S 1 . Other dimensions I for input data N-M Instead of using all multiplication methods of all linear states as the original NG-RC, a sparse connection method is adopted, namely randomly selected input data are multiplied with the original input data in sequence to obtain an additional sparse connection set S 2 。S 1 And S is equal to 2 As a nonlinear eigenvector set inside the reservoir, will combine with the remaining constant and linear terms into O total (t) as a reserve tank.
S2, dividing all regional data groups taking time as sampling labels into two groups, wherein the data groups of most regions are used as training data, and the data groups of few regions are used as target prediction data; the overall data set is divided into two parts in the time dimension, the data of the early most time series are used as training data, and the rest are used as test data; the input data and the output data are subjected to a certain time difference t in the time dimension of training p As the predicted time.
S3, inputting the input sequence in unit time into the optimized NG-RC framework to obtain a linear and nonlinear state vector O of the input sequence total (t)。
S4, utilizing a least square method to combine the state vector with O total With target prediction data I M Performing data matching to obtain an output weight W out 。
S5, outputting weight W out After training, the rest input data I N S3, obtaining a state direction by adopting the same data processing methodQuantity O total 。
S6, the state vector O total Output weight W for training out Multiplication to obtain output predicted data I M 。
The technical scheme of the embodiment is further described in detail below with reference to the attached drawings.
For step S1, the internal structural explanation of step S3 may be referred to in terms of the algorithm structure.
For step S2, the first 90% of N sets of training data in time units is used as training set I N Input data I of last 10% N As a continuous input to the test set. The data, after being input into our novel NG-RC reservoir structure, will generate a linear term S n2 And a nonlinear feature vector set S nl And adding a set of constant terms S c 。
The system realizes the verification of a certain infectious disease development data set in a part of areas. We collected 719 days of succession from 5 months 1 to 4 months 19 days 2022 in 155 countries or regions 2020. All region 155-dimensional data with time 'day' as sampling label are divided into two groups, wherein random 150-group data group is used as training data I N The remaining 5 sets of data are taken as target prediction data I M . The first 560 days of 719 days of data were taken over time as experimental data, where I N The first 500 days are used as training data, and the remaining 30 days are used as test data. Taking time difference t p =21 as the prediction time of the target data, i.e. target prediction data I M The first 21 data points of (1) are discarded, the remainder I M (22:522) target output data as training phase, remaining I M (523:560) as predicted test comparison data. All data were pre-processed at the same time, we used the savgol filter (Python package) to smooth the data.
For step S3, input training data I N Linear and nonlinear combinations are performed, wherein I at a single point in time N Data is 1 x 150 vector, which we extend to state vector O total Is a kind of medium. Wherein the constant term S c Represented by a constant of 1; linear term S n2 Is represented by I N (t) and I N One-dimensional vector of data superposition between (t-1), using I N (t) sequence and I N (t-1) a sequence combination representation; nonlinear feature vector set S nl Representing the algorithm structure versus the linear term S n2 This is similar in part to the high-dimensional nonlinear projection of the reservoir in a conventional RC.
Referring to fig. 1, first, a data preparation layer divides all regional data groups with time T as sampling tags into two groups, wherein 95% of the data groups are used as training data I N 5% of the data sets are taken as target prediction data I M . For our input time series I N At the current time point t, adopting a time overlapping method, adopting the parameter k as a time contact parameter, namely I N (t) and I N K groups of data between (t-k) are used as input data in unit time, so that the relation of the data in the time dimension is enhanced. As shown in FIG. 1, will I N Each item of (t) and the previous instant I N (t-1) into a unified vector, totaling 150×2=300; nonlinear feature vector set S nl Includes I M1 (t) and the previous time unit I M1 (t-1) obtaining a full-connected group by multiplying in turn, and sharing [ (10+1) ×10]2 = 55 items; input data I for other dimensions N-M (t)，I M1 (t) and I N-M (t) and previous time Unit I N-M (t-1) are multiplied in turn to obtain a sparse connected set, and (145×10×2) =2900 items are shared. The three-part addition has 3256 items, namely state vector O total The feature matrix is 500 x 3256 in overall time.
For step S4, the least square method is used: w (W) out ＝YX T (XX T +λⅡ) -1 Fitting output weight W out . Wherein λ is a bias parameter of size 1×10 to avoid overfitting -7 II is an identity matrix. Y represents target prediction data I M X represents a state vector O formed by the input data through the step S2 total . Finally, a weight matrix of 5 x 3256 is output
For step S5, training data I N The number remaining after the step S2According to I N (501:560) similarly utilizes the state vector O of S2 total The generating method obtains a group of 60 x 3256 characteristic matrixes.
For step S6, the feature matrix generated in step S4 is combined with the output weight W generated in step S3 out Multiplying to obtain target prediction data I p . The prediction results are shown in fig. 2, and fig. 2 shows 4 sets of prediction data representing epidemic situation prediction results of 4 regions, namely region a, region B, region C and region D, respectively. It can be seen that the prediction data can show the development trend of epidemic situation about 10 days in the future to a certain extent, wherein the prediction result of the area A can be extended to about 20 days.
The design is realized through a python software design and is deployed on a personal computer based on a 64-bit Windows operating system.
The embodiment refers to an optimized infectious disease prediction method based on next generation reserve pool calculation (Next generation reservoir computing, NG-RC), and firstly, the algorithm is subjected to targeted structure optimization, so that the operation efficiency under the condition of big data input is improved. In the data processing step, the data is divided into a training group and a prediction group, the data has time difference under the same time scale, and after the data is trained, the output weight parameters between the input data and the output data are obtained by utilizing an algorithm. Under continuous input of input data, target data to be predicted is obtained by using the output weight. The present embodiment attempts to overcome the disadvantages of the conventional prediction method that it is severely dependent on a plurality of structural parameters and inherent parameters of the model itself, and to overcome the disadvantage of insufficient data at a point of time. Meanwhile, the infection trend of the target position is obtained by utilizing the infection trend of other areas, so that the aim of really predicting the future infection trend is fulfilled. Finally, through operation optimization of the standard NG-RC, the overall operation efficiency of the system is improved.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to the invention without departing from the principles thereof are intended to be within the scope of the invention as set forth in the following claims.
Claims (4)
1. An infectious disease prediction method based on optimized next generation pool calculation, characterized in that the infectious disease prediction method comprises the following steps:
s1, carrying out internal connection optimization on the basis of an original NG-RC framework structure to obtain an optimized NG-RC framework; specifically, the input dimension is set to be N, the output dimension is set to be M, and the input data is recorded as I N The target predicted output data is denoted as I M The method comprises the steps of carrying out a first treatment on the surface of the At the input data I N Randomly selecting data with the same dimension as the output data, and marking the data as I M1 The method comprises the steps of carrying out a first treatment on the surface of the At unit time t, I M1 (t) and the previous time unit I M1 (t-1) obtaining a full connection group by adopting a way of sequential multiplication; input data I for other dimensions N-M (t)，I M1 (t) and I N-M (t) and previous time Unit I N-M (t-1) multiplying in sequence to obtain a sparse connection set;
s2, collecting infectious disease data of a plurality of areas within a period of time, generating all area data sets taking time as sampling labels, dividing the data sets into two groups, wherein most area data sets are taken as input data I N A few regional data sets as target prediction output data I M The method comprises the steps of carrying out a first treatment on the surface of the Input data I N The method comprises the steps of dividing the data into two parts in the time dimension, taking the data of the early most time sequences as training input data, and taking the rest data as test input data; taking a preset time difference t in the time dimension of training p As a prediction time, target prediction output data I M T is the front of (1) p Discarding the data points of time, and respectively taking the rest data as target fitting data and test comparison data of a training stage;
s3, inputting training input data into the optimized NG-RC framework to obtain a full connection group S corresponding to the training stage 1 And sparse connection set S 2 The method comprises the steps of carrying out a first treatment on the surface of the Will be fully connected to group S 1 And sparse connection set S 2 Combining as a set of nonlinear eigenvectors S inside a reservoir n1 Directing nonlinear characteristics toQuantity set S n1 Sum constant term S c Linear term S n2 Combined into o total A state vector;
s4, using least square method to make state vector O total With target prediction data I M Performing data matching fitting to obtain an output weight W out ；
S5, inputting the test input data into the optimized NG-RC framework to obtain a state vector O total ；
S6, the state vector O total Output weight W for training out Performing dot product multiplication to obtain output predicted data I P ；
In step S3, the state vector O total The calculation formula of (t) is: o (O) total (t)＝[S c ；S l ；S nl ] T (t)；
Wherein the constant term S is represented by a constant 1 c The method comprises the steps of carrying out a first treatment on the surface of the Use I M (t) sequence and I M (t-1) sequence combinations represent a linear term S n2 The method comprises the steps of carrying out a first treatment on the surface of the Use I M (t) sequence and I M (t-1) sequence part combination relation means nonlinear feature vector group S nl Wherein S is nl The method comprises a full connection group and a sparse connection group;
the least square method formula adopted in the step S4 is:
W out ＝YX T (XX T +λII) -1
wherein Y represents target prediction data I N-M X represents a state vector O formed by input data total The method comprises the steps of carrying out a first treatment on the surface of the Lambda is the bias parameter and takes the value of 1 multiplied by 10 -7 II is an identity matrix.
2. The infectious disease prediction method based on optimized next generation reserve pool calculation according to claim 1, wherein in step S2, the process of grouping all regional data groups with time as sampling tags into two groups includes the steps of:
s21, dividing all regional data with time T as sampling label into two groups, wherein 95% of the data groups are used as input data I N Data set of 5% asTarget prediction output data I M ；
S22, input data I N The first 90% on time series is used as training input data, and the last 10% is used as test input data;
s23, for M groups of target prediction output data, adopting a preset time difference t p As a predicted time interval.
3. The infectious disease prediction method based on optimized next generation pool calculation according to claim 1, wherein the outputted prediction data I P The calculation formula of (2) is as follows: i P ＝W out *O total 。
4. The infectious disease prediction method based on optimized next generation pool calculation according to claim 1, wherein in step S2, all data is preprocessed, in particular, data smoothing is performed using savgol filter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310460666.4A CN116168847B (en) | 2023-04-26 | 2023-04-26 | Infectious disease prediction method based on optimized next generation reserve pool calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310460666.4A CN116168847B (en) | 2023-04-26 | 2023-04-26 | Infectious disease prediction method based on optimized next generation reserve pool calculation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116168847A CN116168847A (en) | 2023-05-26 |
| CN116168847B true CN116168847B (en) | 2023-08-11 |
Family
ID=86413641
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310460666.4A Active CN116168847B (en) | 2023-04-26 | 2023-04-26 | Infectious disease prediction method based on optimized next generation reserve pool calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116168847B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108172301A (en) * | 2018-01-31 | 2018-06-15 | 中国科学院软件研究所 | A kind of mosquito matchmaker's epidemic Forecasting Methodology and system based on gradient boosted tree |
| CN111524611A (en) * | 2020-04-24 | 2020-08-11 | 腾讯科技（深圳）有限公司 | Method, device and equipment for constructing infectious disease trend prediction model |
| CN114388138A (en) * | 2022-01-11 | 2022-04-22 | 平安科技（深圳）有限公司 | Artificial intelligence based epidemic situation prediction method, device, equipment and storage medium |
| CN115114979A (en) * | 2022-06-23 | 2022-09-27 | 中国人民解放军国防科技大学 | Infectious disease propagation state prediction method and device based on clustering |
| CN115188480A (en) * | 2022-07-21 | 2022-10-14 | 西南大学 | Infectious disease prediction method based on nano photon reserve pool calculation |
| CN115631869A (en) * | 2022-11-28 | 2023-01-20 | 北京理工大学 | Construction method of infectious disease prediction model |
-
2023
- 2023-04-26 CN CN202310460666.4A patent/CN116168847B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108172301A (en) * | 2018-01-31 | 2018-06-15 | 中国科学院软件研究所 | A kind of mosquito matchmaker's epidemic Forecasting Methodology and system based on gradient boosted tree |
| CN111524611A (en) * | 2020-04-24 | 2020-08-11 | 腾讯科技（深圳）有限公司 | Method, device and equipment for constructing infectious disease trend prediction model |
| CN114388138A (en) * | 2022-01-11 | 2022-04-22 | 平安科技（深圳）有限公司 | Artificial intelligence based epidemic situation prediction method, device, equipment and storage medium |
| CN115114979A (en) * | 2022-06-23 | 2022-09-27 | 中国人民解放军国防科技大学 | Infectious disease propagation state prediction method and device based on clustering |
| CN115188480A (en) * | 2022-07-21 | 2022-10-14 | 西南大学 | Infectious disease prediction method based on nano photon reserve pool calculation |
| CN115631869A (en) * | 2022-11-28 | 2023-01-20 | 北京理工大学 | Construction method of infectious disease prediction model |
Non-Patent Citations (1)
| Title |
|---|
| 重症手足口病早期识别指标研究进展;邓星超;《应用预防医学》;第26卷(第4期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116168847A (en) | 2023-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yu et al. | A novel framework for wind speed prediction based on recurrent neural networks and support vector machine | |
| CN111091045B (en) | Sign language identification method based on space-time attention mechanism | |
| CN107437096B (en) | Image classification method based on parameter efficient depth residual error network model | |
| Chen et al. | ReGAN: A pipelined ReRAM-based accelerator for generative adversarial networks | |
| CN109685819B (en) | Three-dimensional medical image segmentation method based on feature enhancement | |
| CN112001887B (en) | Full convolution genetic neural network method for infant brain medical record image segmentation | |
| Huang et al. | LTNN: A layerwise tensorized compression of multilayer neural network | |
| CN110991317B (en) | Crowd counting method based on multi-scale perspective sensing network | |
| CN109711442B (en) | Unsupervised layer-by-layer generation confrontation feature representation learning method | |
| CN112734748A (en) | Image segmentation system for hepatobiliary and biliary calculi | |
| CN108304912A (en) | A kind of system and method with inhibiting signal to realize impulsive neural networks supervised learning | |
| Zhang et al. | FAT-DeepFFM: Field attentive deep field-aware factorization machine | |
| CN113688765A (en) | Attention mechanism-based action recognition method for adaptive graph convolution network | |
| Chen et al. | CNN-based broad learning with efficient incremental reconstruction model for facial emotion recognition | |
| Chakraborty et al. | Bio-medical image enhancement using hybrid metaheuristic coupled soft computing tools | |
| CN112712170A (en) | Neural morphology vision target classification system based on input weighted impulse neural network | |
| CN113298131B (en) | Attention mechanism-based time sequence data missing value interpolation method | |
| CN116168847B (en) | Infectious disease prediction method based on optimized next generation reserve pool calculation | |
| CN114742210A (en) | Hybrid neural network training method, traffic flow prediction method, apparatus, and medium | |
| CN111652359B (en) | Multiplier array for matrix operations and multiplier array for convolution operations | |
| CN117576402A (en) | Deep learning-based multi-scale aggregation transducer remote sensing image semantic segmentation method | |
| Duggal et al. | Shallow SqueezeNext: An Efficient & Shallow DNN | |
| Xu et al. | Eurnet: Efficient multi-range relational modeling of spatial multi-relational data | |
| CN116152570A (en) | Commodity image classification method based on hierarchical convolutional neural network | |
| Wani et al. | Optimization of deep network models through fine tuning |
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 |