CN113298191A - User behavior identification method based on personalized semi-supervised online federal learning - Google Patents
User behavior identification method based on personalized semi-supervised online federal learning Download PDFInfo
- Publication number
- CN113298191A CN113298191A CN202110766924.2A CN202110766924A CN113298191A CN 113298191 A CN113298191 A CN 113298191A CN 202110766924 A CN202110766924 A CN 202110766924A CN 113298191 A CN113298191 A CN 113298191A
- Authority
- CN
- China
- Prior art keywords
- gradient
- client
- semi
- personalized
- supervised
- 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
Images
Classifications
-
- 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
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of intelligent equipment user behavior recognition, and particularly relates to a user behavior recognition method based on personalized semi-supervised online federal learning. The personalized federal user behavior identification method based on semi-supervised online learning is characterized by comprising the following steps of: determining a client with a label and a client without the label; performing on-line semi-supervised federal learning by using a FedHAR algorithm, and training a generalized neural network model; and carrying out personalized federal fine adjustment on the generalized neural network model to obtain a multi-mode personalized neural network model. The invention provides an individualized user behavior identification method by using a semi-supervised online learning and federal learning framework, which is used for solving the behavior identification problem in a real scene and the privacy problem.
Description
Technical Field
The invention belongs to the technical field of intelligent equipment user behavior recognition, and particularly relates to a user behavior recognition method based on personalized semi-supervised online federal learning.
Background
With the development of sensor technology and the improvement of mobile phone computing capability, the user behavior identification based on the smart phone sensor becomes a research hotspot in recent years. The user behavior identification takes the raw data of a sensor of a mobile phone or wearable equipment as input, and the motion behavior of a user is predicted through an identification algorithm. The system plays an important role in the fields of health and exercise monitoring, user biological feature signatures, urban calculation, assistance of disabled people, old people nursing, indoor positioning and the like. At present, most of sensor-based user behavior recognition research focuses on recognition of simple behaviors, and in many researches, the accuracy rate of recognition of specific simple behaviors can reach more than 95%. The complex behavior recognition task is wide in related range and high in recognition difficulty, and besides various limb behaviors, the complex behavior recognition also relates to recognition of vehicles, recognition of surrounding real environments, indoor positioning and the like. In particular, to enable the user behavior recognition method to be applicable to a real-world scene, there are four challenges to be solved, namely privacy protection, tag scarcity, instantaneity and heterogeneity. The current research on this problem has many deficiencies.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a personalized federal user behavior recognition framework to overcome the obstacles. In particular, the framework utilizes federated learning to train in a distributed environment, where training data can be saved locally to protect the privacy of the user. In addition, for each client without an active label, an algorithm for calculating unsupervised gradients is designed based on consistency training. In addition, an unsupervised gradient aggregation strategy is provided for the problems of concept drift and convergence instability in online learning.
The technical scheme adopted by the invention for solving the technical problems is as follows: a personalized federal user behavior identification method based on semi-supervised online learning comprises the following steps:
the method comprises the following steps: determining a client with a label and a client without the label;
step two: performing on-line semi-supervised federal learning by using a FedHAR algorithm, and training a generalized neural network model;
step three: and carrying out personalized federal fine adjustment on the generalized neural network model to obtain a multi-mode personalized neural network model.
As a preferred mode of the present invention, in step one, any tagged client has a local data set, wherein samples in the data are tagged; any non-label client has a data stream, and the data in the data stream is non-label.
Further, in the second step, the neural network structure is composed of a plurality of attention layers, which are sequentially from bottom to top: the first attention layer fuses time sequence data in a time window of the sensor and carries out weighted average according to the importance degree of the data in the time sequence data; the second layer attention layer is used for carrying out weighted average on the features extracted from the first layer attention layer in the sensors of the same type in a plurality of devices; the third attention layer is used for carrying out weighted average on the characteristics of all the sensors extracted by the second layer and fusing the characteristics into a vector; the obtained vector predicts the label through the output layer.
Further, in the third step, the step of fine tuning the generalized neural network model obtained in the second step by using a PerFedHAR algorithm is as follows:
(1) determining a label-free client m, initializing a fine tuning model into a generalization model: fmSetting the current wheel number to be 1 as F; f represents a generalized neural network model;
(2) if the current number of rounds is less than the set number of rounds RmEntering the next step, otherwise entering the step (9);
(3) transmitting the Current model FmTo all the labeled clients and the unlabeled clients m;
(4) for each labeled client, taking data of one batch from a local database, and calculating a gradient valueFor the unlabeled client m, calculating the gradient by using a gradient calculation method of the unlabeled client and recording the calculated gradient as
(5) After receiving all the gradients transmitted from the labeled clients and the gradient from the unlabeled client m, the server averages all the labeled gradients; average of all labeled gradients
(6) Weighting the supervised gradient and the semi-supervised gradient to calculate the final gradient
(7) Using gradient pairs FmUpdating and adding one to the number of the rounds;
(8) entering (2);
and (6) ending.
Further, the gradient calculation method of the untagged client is as follows:
(1) receiving a generalized neural network model F transmitted from a server side;
(2) taking the last saved sample x from storage1If not, obtaining a sample from the current data stream and assigning the sample to x1;
(5) taking one sample from the current data stream and recording the sample as x2;
(6) Predicting x using model F pairs1、x2Before, inMeasured values are y respectively1、y2;
(8) Entering (4);
Compared with the prior art, the invention has the following advantages and beneficial effects:
the invention solves the behavior recognition problem in a real scene by using a personalized federal user behavior recognition method based on semi-supervised online learning. The existing method usually trains data in a centralized batch processing mode, which not only leaks the privacy of users, but also requires a very large storage space to store all sensor data. Furthermore, the rarity of action labels is also an extremely serious problem. Finally, although the trained generalized model has good performance on all clients, the generalized model cannot achieve the best performance on a specific client due to heterogeneity of behavior patterns of individuals. The invention provides an individualized user behavior identification method by using a semi-supervised online learning and federal learning framework. In summary, the above technical solutions conceived by the present invention have the following technical features and beneficial effects compared with the prior art:
a general FedHAR framework is provided to solve four problems of privacy protection, label scarcity, instantaneity and heterogeneity in action recognition. FedHAR can train a generalized federated model with only a small number of labeled clients with limited samples, with competitive performance, and using the large amount of real-time flow knowledge data generated by unlabeled clients.
A new algorithm for calculating unsupervised gradients under a consistency training proposition is designed. We have devised an unsupervised gradient aggregation strategy to overcome the concept drift and convergence instability problems in online learning. Where semi-supervised learning loss is calculated from gradients from all tagged and untagged clients.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention.
FIG. 2 is a flow chart of a generalized model trained using the FedHAR algorithm;
FIG. 3 is a process for fine-tuning a particular untagged client using the PerFedHAR algorithm;
FIG. 4 is a flow chart of a gradient computation method for a label-free client;
FIG. 5 is a schematic diagram of a multi-modal hierarchical attention neural network model proposed by the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Embodiment 1 a personalized federal user behavior recognition method based on semi-supervised online learning provided in this embodiment is as shown in fig. 1, and mainly includes the following steps:
firstly, the method comprises the following steps: and determining the client with the label and the client without the label, and performing preparation work.
In this step, information of each client in federal learning is first determined. In this embodiment, the clients are divided into two types, one is a client having a data set with a sample with a tag (hereinafter referred to as a tagged client), and the other is a client having an untagged data stream (hereinafter referred to as an untagged client). In any tagged client, there is a local data set in which samples in the data are tagged. And any non-label client has a data stream, and the data in the data stream is non-label. This accords with real life scene very much in fact, and the smart machine that people wore can produce the no label sensor data of continuous in real time.
II, secondly: and (5) carrying out on-line semi-supervised federal learning by using a FedHAR algorithm, and training a generalized model. The training process is shown in fig. 2, and specifically includes:
(1) the neural network F is randomly initialized. The current number of wheels is set to 1.
(2) And judging whether the current round number reaches the specified round number R. If the value is less than R, the execution is continued. Otherwise, go to step (12)
(3) Random selection of MbA non-label client (denoted as B)M) And sending the network F to the selected non-labeled client and all the labeled clients.
(4) For any client with a label, randomly taking out a piece of data from a local database, and then calculating the gradientFor any unlabeled client m, the gradient is calculated using the gradient calculation method of the unlabeled client (see below)Then, the agg gradients are averaged, and the average gradient of the client is obtained
(5) And when the server receives all the gradients transmitted from the labeled client and calculates all the average gradients of the unlabeled clients, averaging all the labeled gradients and the unlabeled gradients. Average of all labeled gradientsAnd the mean value of all unlabeled gradients is recorded
(6) If the number of wheels is larger than the set over-parameter rλThen (7) is entered, otherwise (8) is entered.
(7) The semi-supervised weight λ' is set to the hyperparameter λ, and (9) is entered.
(9) Weighting the supervised gradient and the semi-supervised gradient to calculate the final gradient
(10) F is updated using the gradient and the number of rounds is incremented by one.
(11) Enter (2)
(12) And finishing to obtain a generalized model F.
This step mainly explains from the server side how to train out a generalized neural network model.
Thirdly, the method comprises the following steps: and fine-tuning each label-free client by using the trained generalized model, and fine-tuning a most suitable model for each label-free client.
Fig. 3 illustrates a process of fine-tuning a generalized neural network model of a specific unlabeled client by using a PerFedHAR algorithm, which includes the following specific steps:
(1) determining a specific label-free client m, and initializing a fine tuning model into a generalization model: fmSet the current number of wheels to 1.
(2) If the current number of rounds is less than the set number of rounds RmAnd entering the next step, and otherwise entering the step (9).
(3) Transmitting the Current model FmTo all tagged and untagged clients m.
(4) And for each labeled client, taking out data of one batch from the local database, and calculating a gradient value. For the untagged client m, the gradient is calculated by using the gradient calculation method (see below) of the untagged client
(5) Receiving by serverAfter all the gradients passed from the tagged clients and the gradient from the untagged client m, all the tagged gradients are averaged. The average of all labeled gradients was recorded as:
(6) weighting the supervised gradient and the semi-supervised gradient to calculate the final gradient
(7) Using gradient pairs FmUpdating and adding one to the number of rounds.
(8) And (3) entering into (2).
(9) And (6) ending.
The flow of the gradient calculation method of the label-free client is shown in fig. 4, and the specific steps are as follows:
(1) and receiving the model F transmitted from the server side.
(2) Taking the last saved sample x from storage1If not, obtaining a sample from the current data stream and assigning the sample to x1。
(5) Taking one sample from the current data stream and recording the sample as x2。
(6) Predicting x using model F pairs1、x2Predicted values are y respectively1、y2。
(8) Enter (4)
The step mainly explains how to fine-tune the trained generalization model from the server side. The network structure of the trained generalized neural network model F is shown in fig. 5. The neural network model is composed of a plurality of attention layers, and data of a plurality of modes can be organically fused. From bottom to top, the lowest first layer attention layer is to fuse the time series data in a time window of the sensor and perform weighted average according to the importance degree of the data in the time series data. The second layer of attention is a weighted average of the features extracted from the first layer of attention in the same type of sensor in the plurality of devices. The third layer of attention layer is to carry out weighted average on the characteristics of all the types of sensors extracted from the second layer and to combine the characteristics into a vector. Finally this vector predicts the label through the output layer.
Claims (5)
1. A personalized federal user behavior identification method based on semi-supervised online learning is characterized by comprising the following steps:
the method comprises the following steps: determining a client with a label and a client without the label;
step two: performing on-line semi-supervised federal learning by using a FedHAR algorithm, and training a generalized neural network model;
step three: and carrying out personalized federal fine adjustment on the generalized neural network model to obtain a multi-mode personalized neural network model.
2. The method for personalized federal user behavior identification based on semi-supervised online learning as claimed in claim 1, wherein in the step one, any one of the labeled clients has a local data set, wherein the samples in the data are labeled; any non-label client has a data stream, and the data in the data stream is non-label.
3. The method for personalized federal user behavior recognition based on semi-supervised online learning as claimed in claim 1, wherein in the second step, the neural network structure is composed of a plurality of attention layers, which are sequentially from bottom to top: the first attention layer fuses time sequence data in a time window of the sensor and carries out weighted average according to the importance degree of the data in the time sequence data; the second layer attention layer is used for carrying out weighted average on the features extracted from the first layer attention layer in the sensors of the same type in a plurality of devices; the third attention layer is used for carrying out weighted average on the characteristics of all the sensors extracted by the second layer and fusing the characteristics into a vector; the obtained vector predicts the label through the output layer.
4. The method for identifying personalized federal user behaviors based on semi-supervised online learning as claimed in claim 1, wherein in the third step, the step of fine tuning the generalized neural network model obtained in the second step by using a PerFedHAR algorithm is as follows:
(1) determining a label-free client m, initializing a fine tuning model into a generalization model: fmSetting the current wheel number to be 1 as F; f represents a generalized neural network model;
(2) if the current number of rounds is less than the set number of rounds RmEntering the next step, otherwise entering the step (9);
(3) transmitting the Current model FmTo all the labeled clients and the unlabeled clients m;
(4) for each labeled client, taking data of one batch from a local database, and calculating a gradient valueFor the unlabeled client m, calculating the gradient by using a gradient calculation method of the unlabeled client and recording the calculated gradient as
(5) After receiving all the gradients transmitted from the labeled clients and the gradient from the unlabeled client m, the server averages all the labeled gradients; average of all labeled gradients
(6) Weighting the supervised gradient and the semi-supervised gradient to calculate the final gradient
(7) Using gradient pairs FmUpdating and adding one to the number of the rounds;
(8) entering (2);
(9) and (6) ending.
5. The method for personalized federal user behavior recognition based on semi-supervised online learning as claimed in claim 4, wherein the gradient calculation method of the untagged client comprises the following steps:
(1) receiving a generalized neural network model F transmitted from a server side;
(2) taking the last saved sample x from storage1If not, obtaining a sample from the current data stream and assigning the sample to x1;
(5) taking one sample from the current data stream and recording the sample as x2;
(6) Predicting x using model F pairs1、x2Predicted values are y respectively1、y2;
(8) Entering (4);
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2021103566641 | 2021-04-01 | ||
CN202110356664 | 2021-04-01 |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113298191A true CN113298191A (en) | 2021-08-24 |
CN113298191B CN113298191B (en) | 2022-09-06 |
Family
ID=77330767
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110766924.2A Active CN113298191B (en) | 2021-04-01 | 2021-07-07 | User behavior identification method based on personalized semi-supervised online federal learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113298191B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113705712A (en) * | 2021-09-02 | 2021-11-26 | 广州大学 | Network traffic classification method and system based on federal semi-supervised learning |
CN113762526A (en) * | 2021-09-07 | 2021-12-07 | 京东科技信息技术有限公司 | Federal learning method, hierarchical network system, storage medium and electronic equipment |
CN116842577A (en) * | 2023-08-28 | 2023-10-03 | 杭州海康威视数字技术股份有限公司 | Federal learning model poisoning attack detection and defense method, device and equipment |
CN117094381A (en) * | 2023-08-21 | 2023-11-21 | 哈尔滨工业大学 | Multi-mode federal collaboration method taking high-efficiency communication and individuation into consideration |
CN117910539A (en) * | 2024-03-19 | 2024-04-19 | 电子科技大学 | Household characteristic recognition method based on heterogeneous semi-supervised federal learning |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110941734A (en) * | 2019-11-07 | 2020-03-31 | 南京理工大学 | Depth unsupervised image retrieval method based on sparse graph structure |
CN111275207A (en) * | 2020-02-10 | 2020-06-12 | 深圳前海微众银行股份有限公司 | Semi-supervision-based horizontal federal learning optimization method, equipment and storage medium |
CN111401552A (en) * | 2020-03-11 | 2020-07-10 | 浙江大学 | Federal learning method and system based on batch size adjustment and gradient compression rate adjustment |
CN111832503A (en) * | 2020-07-20 | 2020-10-27 | 中国人民武装警察部队工程大学 | Distracted driving identification method based on federal learning |
CN112052960A (en) * | 2020-09-07 | 2020-12-08 | 深圳前海微众银行股份有限公司 | Longitudinal federal modeling method, device, equipment and computer readable storage medium |
CN112287089A (en) * | 2020-11-23 | 2021-01-29 | 腾讯科技(深圳)有限公司 | Classification model training and automatic question-answering method and device for automatic question-answering system |
CN112507219A (en) * | 2020-12-07 | 2021-03-16 | 中国人民大学 | Personalized search system based on federal learning enhanced privacy protection |
-
2021
- 2021-07-07 CN CN202110766924.2A patent/CN113298191B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110941734A (en) * | 2019-11-07 | 2020-03-31 | 南京理工大学 | Depth unsupervised image retrieval method based on sparse graph structure |
CN111275207A (en) * | 2020-02-10 | 2020-06-12 | 深圳前海微众银行股份有限公司 | Semi-supervision-based horizontal federal learning optimization method, equipment and storage medium |
CN111401552A (en) * | 2020-03-11 | 2020-07-10 | 浙江大学 | Federal learning method and system based on batch size adjustment and gradient compression rate adjustment |
CN111832503A (en) * | 2020-07-20 | 2020-10-27 | 中国人民武装警察部队工程大学 | Distracted driving identification method based on federal learning |
CN112052960A (en) * | 2020-09-07 | 2020-12-08 | 深圳前海微众银行股份有限公司 | Longitudinal federal modeling method, device, equipment and computer readable storage medium |
CN112287089A (en) * | 2020-11-23 | 2021-01-29 | 腾讯科技(深圳)有限公司 | Classification model training and automatic question-answering method and device for automatic question-answering system |
CN112507219A (en) * | 2020-12-07 | 2021-03-16 | 中国人民大学 | Personalized search system based on federal learning enhanced privacy protection |
Non-Patent Citations (2)
Title |
---|
KONSTANTIN SOZINOV ET AL.: ""Human Activity Recognition Using Federated Learning"", 《2018 IEEE INTL CONF ON PARALLEL & DISTRIBUTED PROCESSING WITH APPLICATIONS, UBIQUITOUS COMPUTING & COMMUNICATIONS, BIG DATA & CLOUD COMPUTING, SOCIAL COMPUTING & NETWORKING, SUSTAINABLE COMPUTING & COMMUNICATIONS》 * |
王蓉 等: ""基于联邦学习和卷积神经网络的入侵检测方法"", 《信息网络安全》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113705712A (en) * | 2021-09-02 | 2021-11-26 | 广州大学 | Network traffic classification method and system based on federal semi-supervised learning |
CN113705712B (en) * | 2021-09-02 | 2024-01-05 | 广州大学 | Network traffic classification method and system based on federal semi-supervised learning |
CN113762526A (en) * | 2021-09-07 | 2021-12-07 | 京东科技信息技术有限公司 | Federal learning method, hierarchical network system, storage medium and electronic equipment |
CN113762526B (en) * | 2021-09-07 | 2024-02-09 | 京东科技信息技术有限公司 | Federal learning method, hierarchical network system, storage medium and electronic device |
CN117094381A (en) * | 2023-08-21 | 2023-11-21 | 哈尔滨工业大学 | Multi-mode federal collaboration method taking high-efficiency communication and individuation into consideration |
CN117094381B (en) * | 2023-08-21 | 2024-04-12 | 哈尔滨工业大学 | Multi-mode federal collaboration method taking high-efficiency communication and individuation into consideration |
CN116842577A (en) * | 2023-08-28 | 2023-10-03 | 杭州海康威视数字技术股份有限公司 | Federal learning model poisoning attack detection and defense method, device and equipment |
CN116842577B (en) * | 2023-08-28 | 2023-12-19 | 杭州海康威视数字技术股份有限公司 | Federal learning model poisoning attack detection and defense method, device and equipment |
CN117910539A (en) * | 2024-03-19 | 2024-04-19 | 电子科技大学 | Household characteristic recognition method based on heterogeneous semi-supervised federal learning |
CN117910539B (en) * | 2024-03-19 | 2024-05-31 | 电子科技大学 | Household characteristic recognition method based on heterogeneous semi-supervised federal learning |
Also Published As
Publication number | Publication date |
---|---|
CN113298191B (en) | 2022-09-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113298191B (en) | User behavior identification method based on personalized semi-supervised online federal learning | |
CN106600059B (en) | Intelligent power grid short-term load prediction method based on improved RBF neural network | |
Kong et al. | Integrated generative model for industrial anomaly detection via bidirectional LSTM and attention mechanism | |
CN109858390A (en) | The Activity recognition method of human skeleton based on end-to-end space-time diagram learning neural network | |
CN114220271A (en) | Traffic flow prediction method, equipment and storage medium based on dynamic space-time graph convolution cycle network | |
CN111737592B (en) | Recommendation method based on heterogeneous propagation collaborative knowledge sensing network | |
Bu | An efficient fuzzy c-means approach based on canonical polyadic decomposition for clustering big data in IoT | |
CN113378074A (en) | Social network user trajectory analysis method based on self-supervision learning | |
CN111178288B (en) | Human body posture recognition method and device based on local error layer-by-layer training | |
CN113313232A (en) | Functional brain network classification method based on pre-training and graph neural network | |
Wan et al. | Network traffic prediction based on LSTM and transfer learning | |
CN116596151B (en) | Traffic flow prediction method and computing device based on time-space diagram attention | |
CN112766603A (en) | Traffic flow prediction method, system, computer device and storage medium | |
CN116862024A (en) | Credible personalized federal learning method and device based on clustering and knowledge distillation | |
CN113297936A (en) | Volleyball group behavior identification method based on local graph convolution network | |
CN116052254A (en) | Visual continuous emotion recognition method based on extended Kalman filtering neural network | |
CN114708637A (en) | Face action unit detection method based on meta-learning | |
CN107341471A (en) | A kind of Human bodys' response method based on Bilayer condition random field | |
CN114372680A (en) | Spatial crowdsourcing task allocation method based on worker loss prediction | |
CN117116048A (en) | Knowledge-driven traffic prediction method based on knowledge representation model and graph neural network | |
CN116861923A (en) | Multi-view unsupervised graph contrast learning model construction method, system, computer, storage medium and application | |
CN116523002A (en) | Method and system for predicting dynamic graph generation countermeasure network track of multi-source heterogeneous data | |
Christoforidis et al. | Recommending points of interest in LBSNs using deep learning techniques | |
CN113010772B (en) | Data processing method, related equipment and computer readable storage medium | |
CN115510757A (en) | Design method for long-time sequence prediction based on gated convolution and time attention mechanism |
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 |