CN110880019A - Method for adaptively training target domain classification model through unsupervised domain - Google Patents
Method for adaptively training target domain classification model through unsupervised domain Download PDFInfo
- Publication number
- CN110880019A CN110880019A CN201911045261.4A CN201911045261A CN110880019A CN 110880019 A CN110880019 A CN 110880019A CN 201911045261 A CN201911045261 A CN 201911045261A CN 110880019 A CN110880019 A CN 110880019A
- Authority
- CN
- China
- Prior art keywords
- domain
- target domain
- centroid
- source
- source domain
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Abstract
The invention discloses a method for adaptively training a classification model of a target domain through an unsupervised domain, which comprises the following steps: extracting features of batch image data input by a source domain and a target domain through a standard convolution network, and then constructing an example graph corresponding to the source domain and the target domain by combining with the initially set centroid features; after the node matrix in the example graph sequentially passes through the source domain classifier and the graph convolution network, category centroid characteristics corresponding to a source domain and a target domain are generated; using a class centroid alignment mechanism to constrain class centroids from different domains in each layer of the graph convolution network, so that the class centroids of different domains are gradually close along with iterative training; moreover, a antagonism alignment mechanism guided by the mass center is used, and the mass center automatically generated by all categories is used as the global domain statistical information to guide each batch of image data to participate in antagonism training; and finally obtaining a classification model effective in the target domain through iterative training. The method has good generalization and the trained classification model has high classification accuracy.
Description
Technical Field
The invention relates to the technical field of image classification, in particular to a method for adaptively training a target domain classification model through an unsupervised domain.
Background
Unsupervised domain adaptation can utilize existing source domain tagged data and network models and associated target domain untagged data to learn to derive a network model suitable for target domain data classification.
Conventional unsupervised domain adaptation methods typically use measures such as correlation distance metrics to align the data distribution of the source domain and the target domain output by the deep network. In recent years, many adversarial domain adaptation methods have been proposed and achieved remarkable results, and most of these methods are based on generation of an adversarial network. The method mainly comprises the steps of training a discriminator to discriminate whether a sampling feature is from a source domain or a target domain, and simultaneously training a feature extractor to deceive the discriminator, so that the feature distributions of the source domain and the target domain are aligned and cannot be distinguished.
Most of these methods focus on measuring domain differences at the domain level, without distinguishing whether samples from two domains are aligned according to the category to which they belong. Even if the global statistics are completely mixed up, the difference between the source domain and the target domain is not necessarily reduced, and even different types of samples are mixed together, so that the classification effect is still to be improved.
Disclosure of Invention
The invention aims to provide a method for adaptively training a target domain classification model through an unsupervised domain, which has good generalization and the classification accuracy of the trained classification model is higher.
The purpose of the invention is realized by the following technical scheme:
a method of training a target domain classification model through unsupervised domain adaptation, comprising:
extracting image characteristics of batch image data input by a source domain and a target domain through a standard convolution network, and then constructing an example graph corresponding to the source domain and the target domain by combining with the mass center characteristics set by initialization; after the node matrix in the example graph sequentially passes through the source domain classifier and the graph convolution network, the class centroid characteristics corresponding to the source domain and the target domain are updated;
using a class centroid feature alignment mechanism to constrain class centroid features from different domains in each layer of the graph convolution network, so that the class centroid features of different domains gradually approach along with iterative training;
moreover, a antagonism alignment mechanism guided by the centroid is used, and the centroid characteristics automatically generated by all categories are used as domain global statistical information to guide each batch of image data to participate in antagonism training;
and finally obtaining a classification model effective in the target domain through iterative training.
According to the technical scheme provided by the invention, the network can be trained in an end-to-end mode to automatically learn the class centroid without depending on specific prior knowledge, so that the method has better generalization; according to the method, the average classification accuracy rate is improved by 1-2% on a plurality of data sets, and the convergence rate of the model classification accuracy rate is higher.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are 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 to obtain other drawings based on the drawings without creative efforts.
Fig. 1 is a block diagram of a method for training a target domain classification model through unsupervised domain adaptation according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention are 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 embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention provides a method for training a target domain classification model through unsupervised domain adaptation, the unsupervised domain adaptation related by the method is based on graph convolution category perception structure modeling, as shown in figure 1, the unsupervised domain adaptation is an integral framework of a related method, and the training process mainly comprises the following parts:
1. and automatically generating class centroid characteristics.
Extracting image characteristics of batch image data input by a source domain and a target domain through a standard convolution network, and then constructing an example graph corresponding to the source domain and the target domain by combining with initially set class centroid characteristics; the example graph comprises a node matrix and a corresponding adjacency matrix, wherein the node matrix is composed of a series of characteristic nodes, and after the node matrix sequentially passes through the source domain classifier and the graph convolution network, the class centroid characteristics corresponding to the source domain and the target domain are updated.
In an embodiment of the present invention, the batch image data input by the source domain and the target domain includes: source domain data with a label and target domain data without a label.
Extracting image features through a standard convolution network, and then constructing a node matrix corresponding to a source domain and a target domain by combining with the initially set centroid features to represent as follows:
in the above formula, AlexNet (X)batch) Representing extraction of batch image data X using AlexNet Standard convolutional networkbatchThe features of (1); c represents the category centroid characteristics of the initialization setting;representing a characteristic concatenation; s, T, corresponding to the source and target domains, respectively, i.e.A matrix of nodes corresponding to the source domain,a node matrix corresponding to the target domain.
As shown in FIG. 1, wherein XS、XTFeatures of the extracted source domain and target domain images.
Predicting a node matrix using a source domain classifierSoft label of each characteristic nodeTherefore, the weights of the disconnected edges of the corresponding characteristic nodes and the corresponding adjacent matrixes are constructed according to the similarity among the characteristic nodesExpressed as:
in the above formula, ═ S.T,respectively are an adjacent matrix corresponding to the node matrix of the source domain and a soft label of the source domain characteristic node;respectively is an adjacent matrix corresponding to the node matrix of the target domain and a soft label of the characteristic node of the target domain.
The node matrix and the corresponding adjacent matrix obtained in the above way form a complete example graph.
Then, the node matrixes of the source domain and the target domain respectively pass through a graph convolution network, and the related operation is expressed as follows:
in the above formula, the first and second carbon atoms are,correspondingly representing the output results of the ith layer and the l +1 layer of the graph convolution network;a degree matrix with ith row and ith column elements Is a contiguous matrixRow i and column j, similarly, T corresponds to source and destination domains, W(l+1)The learnable parameters of the l +1 th layer of graph convolution are represented, the parameters in the graph convolution network corresponding to the source domain and the target domain are shared, and sigma represents an activation function.
And transmitting the feature information by using the relationship between the feature nodes, updating the node features, automatically generating new class centroid features and updating the original class centroid features. While the two alignment mechanisms described below for class centroid features will constrain the features to be semantically rich.
In the embodiment of the present invention, since the image samples are to be classified accurately, for each class, there exists a class centroid feature in the feature space of the features of all the image samples to represent all the image samples of the class. Class centroid features do not actually exist inIn the conventional method, the centroid feature is represented by averaging the same-class sample features, in the embodiment of the invention, the class centroid feature is automatically learned by using a graph convolution network, as shown in fig. 1, C is shown on the left side and the right sideSAnd CTBoth represent the class centroid characteristics generated by the source domain and the target domain, with the difference being that the left side is the pre-update class centroid characteristic and the right side is the post-update class centroid characteristic. As will be understood by those skilled in the art, the class centroid feature before updating is the class centroid feature set by the initialization, or the class centroid feature obtained after the last operation.
2. Class centroid feature alignment mechanism.
In practice, the conventional method focuses on modeling class-level information, so that impressive effects are obtained, and the importance of the class-level information is further emphasized. In order to ensure that features from the same class in different domains are mapped to adjacent locations, in an embodiment of the invention, a class centroid feature alignment mechanism is designed to model unsupervised domain adapted class information by constraining class centroid features from different domains in each layer of the graph convolution network such that the class centroid features of different domains come closer together as iterative training progresses, in such a way as to encode the learned features. Thus, samples belonging to the same class can be embedded in nearby locations in the feature space.
The penalty function for the class centroid feature alignment mechanism is:
wherein K represents the number of categories, phi represents the distance metric function, K is the category number, CSAnd CTRespectively representing class centroid characteristics of the source domain and the target domain.
3. Centroid directed antagonism alignment mechanism.
In the embodiment of the invention, a centroid-guided antagonism alignment mechanism is used, and the centroid characteristics automatically generated by all categories are used as domain global statistical information to guide each batch of image data to participate in antagonism training, so that the influence of local information interference discriminators on domain global distribution judgment brought by batch sample input is relieved. The participation of class centroid features may improve training efficiency and adaptability.
The loss function for the centroid directed antagonistic alignment mechanism is:
wherein, χS={x|x∈DS},χT={x|x∈DTRespectively representing source domain and target domain images, CSAnd CTClass centroid features representing the source domain and the target domain, respectively, D representing the discriminator, G representing the feature extractor,representing a series of features.
As shown in fig. 1, the feature extractor is composed of a standard convolution network AlexNet and a graph convolution network.
As shown in fig. 1, the classifier F is a task-specific classifier (training classifier F for short) updated with training for the domain-adapted target, and after the training of the classification model is completed, the training classifier F is used to classify the target domain picture and output a classification score. The discriminator D is used in the training phase, and its main role is to determine whether the image is from the source domain or the target domain according to the input features and the class centroid features, which can be expressed as: d (Concat ([ X ]S,CS],axis=0)),D(Concat([XT,CT]Axis ═ 0)), Concat denotes a splicing operation, and axis ═ 0 denotes that Concat was performed on the 0 th dimension.
The antagonism alignment mechanism designed by the scheme of the embodiment of the invention has participation of the class centroid features, thereby helping the alignment of the class centroid features. In turn, the class centroid feature alignment mechanism constrains learning more semantic class centroid features to help guide the antagonism alignment mechanism. Therefore, the two mechanisms can be mutually strengthened, so that the classification model can be trained in an end-to-end mode without depending on human priors.
Through the mode, the defined loss function is combined for continuous iterative training, and finally the classification model effective in the target domain can be obtained. The classification model obtained by training can effectively classify the input target domain images, and the classification accuracy is high. In the testing stage, the target field image chiTInputting a trained classification model, extracting features through a feature extractor G, namely AlexNet, and combining the training to obtain the target domain centroid featuresFurther extracted feature G (χ) by GCNT) I.e. byExtracting the final characteristic G (χ)T) And inputting the result into a training classifier F to obtain a classification result.
The scheme of the embodiment of the invention can be applied to the pre-classification of large-data-scale label-free images. In practice, the image data processing system can be installed on a working computer in a software mode to perform real-time classified display of small image data batches, and can also be installed on a large server to process large image data batches.
Compared with the prior art, the technical effects are mainly obtained as follows:
the network can be trained in an end-to-end mode to automatically learn the class centroid characteristics without depending on specific prior knowledge of human, so that the method has better generalization; according to the method, the average classification accuracy rate is improved by 1-2% on a plurality of data sets, and the convergence rate of the model classification accuracy rate is higher.
Through the above description of the embodiments, it is clear to those skilled in the art that the above embodiments can be implemented by software, and can also be implemented by software plus a necessary general hardware platform. With this understanding, the technical solutions of the embodiments can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a usb disk, a removable hard disk, etc.), and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods according to the embodiments of the present invention.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (6)
1. A method for training a classification model of a target domain through unsupervised domain adaptation, comprising:
extracting image characteristics of batch image data input by a source domain and a target domain through a standard convolution network, and then constructing an example graph corresponding to the source domain and the target domain by combining with the mass center characteristics set by initialization; after the node matrix in the example graph sequentially passes through the source domain classifier and the graph convolution network, the class centroid characteristics corresponding to the source domain and the target domain are updated;
using a class centroid feature alignment mechanism to constrain class centroid features from different domains in each layer of the graph convolution network, so that the class centroid features of different domains gradually approach along with iterative training;
moreover, a antagonism alignment mechanism guided by the centroid is used, and the centroid characteristics automatically generated by all categories are used as domain global statistical information to guide each batch of image data to participate in antagonism training;
and finally obtaining a classification model effective in the target domain through iterative training.
2. The method of claim 1, wherein the batch of image data input by the source domain and the target domain comprises: source domain data with a label and target domain data without a label.
3. The method of claim 1, wherein the extracting image features through a standard convolutional network and then constructing the instance graph corresponding to the source domain and the target domain in combination with the initially set centroid features comprises:
extracting image features, and constructing a node matrix corresponding to a source domain and a target domain by combining with the initially set centroid features:
in the above formula, AlexNet (X)batch) Representing extraction of batch image data X using AlexNet Standard convolutional networkbatchThe features of (1); c represents the category centroid characteristics of the initialization setting;representing a characteristic concatenation; s, T, corresponding to the source and target domains, respectively, i.e.A matrix of nodes corresponding to the source domain,a node matrix corresponding to the target domain;
predicting node matrices by source domain classifierSoft label of each characteristic nodeTherefore, the weights of the disconnected edges of the corresponding characteristic nodes and the corresponding adjacent matrixes are constructed according to the similarity among the characteristic nodesExpressed as:
wherein, S, T,respectively are an adjacent matrix corresponding to the node matrix of the source domain and a soft label of the source domain characteristic node;respectively are an adjacent matrix corresponding to the node matrix of the target domain and a soft label of the characteristic node of the target domain;
the node matrix obtained in the above manner and the corresponding adjacency matrix form an example graph.
4. The method of claim 3, wherein the node matrices of the source domain and the target domain are each represented by a graph convolution network, and the correlation operation is represented as:
in the above formula, the first and second carbon atoms are,correspondingly representing the output results of the ith layer and the l +1 layer of the graph convolution network;a degree matrix with ith row and ith column elements Is a contiguous matrixRow ith and column jth element, W(l+1)The learnable parameters of the l +1 th layer of graph convolution are represented, the parameters in the graph convolution network corresponding to the source domain and the target domain are shared, and sigma represents an activation function.
5. The method of claim 1, wherein the loss function of the class centroid feature alignment mechanism is:
wherein K represents the number of categories, phi represents the distance metric function, K is the category number, CSAnd CTRespectively representing class centroid characteristics of the source domain and the target domain.
6. The method of claim 1, wherein the loss function of the centroid-directed antagonism alignment mechanism is:
wherein the content of the first and second substances,representing source and target domain images, respectively, CSAnd CTRespectively representing the class centroid characteristics of a source domain and a target domain, D representing a discriminator, G representing a characteristic extractor, and consisting of a standard convolution network and a graph convolution network,representing a series of features.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911045261.4A CN110880019B (en) | 2019-10-30 | 2019-10-30 | Method for adaptively training target domain classification model through unsupervised domain |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911045261.4A CN110880019B (en) | 2019-10-30 | 2019-10-30 | Method for adaptively training target domain classification model through unsupervised domain |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110880019A true CN110880019A (en) | 2020-03-13 |
CN110880019B CN110880019B (en) | 2022-07-12 |
Family
ID=69728198
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911045261.4A Active CN110880019B (en) | 2019-10-30 | 2019-10-30 | Method for adaptively training target domain classification model through unsupervised domain |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110880019B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111612051A (en) * | 2020-04-30 | 2020-09-01 | 杭州电子科技大学 | Weak supervision target detection method based on graph convolution neural network |
CN112629863A (en) * | 2020-12-31 | 2021-04-09 | 苏州大学 | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions |
CN113011456A (en) * | 2021-02-05 | 2021-06-22 | 中国科学技术大学 | Unsupervised domain adaptation method based on class adaptive model for image classification |
CN113128667A (en) * | 2021-04-02 | 2021-07-16 | 中国科学院计算技术研究所 | Cross-domain self-adaptive graph convolution balance migration learning method and system |
CN113255823A (en) * | 2021-06-15 | 2021-08-13 | 中国人民解放军国防科技大学 | Unsupervised domain adaptation method and unsupervised domain adaptation device |
CN113436197A (en) * | 2021-06-07 | 2021-09-24 | 华东师范大学 | Domain-adaptive unsupervised image segmentation method based on generation of confrontation and class feature distribution |
CN113688867A (en) * | 2021-07-20 | 2021-11-23 | 广东工业大学 | Cross-domain image classification method |
CN113723917A (en) * | 2021-08-24 | 2021-11-30 | 中国人民解放军32382部队 | Association construction method and device of instrument management standard and instrument technical standard |
CN113743523A (en) * | 2021-09-13 | 2021-12-03 | 西安建筑科技大学 | Visual multi-feature guided construction waste fine classification method |
CN114048546A (en) * | 2021-11-17 | 2022-02-15 | 大连理工大学 | Graph convolution network and unsupervised domain self-adaptive prediction method for residual service life of aircraft engine |
US11797611B2 (en) | 2021-07-07 | 2023-10-24 | International Business Machines Corporation | Non-factoid question answering across tasks and domains |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108062753A (en) * | 2017-12-29 | 2018-05-22 | 重庆理工大学 | The adaptive brain tumor semantic segmentation method in unsupervised domain based on depth confrontation study |
CN108288051A (en) * | 2018-02-14 | 2018-07-17 | 北京市商汤科技开发有限公司 | Pedestrian identification model training method and device, electronic equipment and storage medium again |
EP3385887A1 (en) * | 2017-04-08 | 2018-10-10 | INTEL Corporation | Sub-graph in frequency domain and dynamic selection of convolution implementation on a gpu |
CN109753992A (en) * | 2018-12-10 | 2019-05-14 | 南京师范大学 | The unsupervised domain for generating confrontation network based on condition adapts to image classification method |
CN109948648A (en) * | 2019-01-31 | 2019-06-28 | 中山大学 | A kind of multiple target domain adaptive migration method and system based on member confrontation study |
CN109948741A (en) * | 2019-03-04 | 2019-06-28 | 北京邮电大学 | A kind of transfer learning method and device |
CA3002100A1 (en) * | 2018-04-18 | 2019-10-18 | Element Ai Inc. | Unsupervised domain adaptation with similarity learning for images |
US20190325299A1 (en) * | 2018-04-18 | 2019-10-24 | Element Ai Inc. | Unsupervised domain adaptation with similarity learning for images |
CN111222471A (en) * | 2020-01-09 | 2020-06-02 | 中国科学技术大学 | Zero sample training and related classification method based on self-supervision domain perception network |
CN111259720A (en) * | 2019-10-30 | 2020-06-09 | 北京中科研究院 | Unsupervised pedestrian re-identification method based on self-supervision agent feature learning |
CN111340021A (en) * | 2020-02-20 | 2020-06-26 | 中国科学技术大学 | Unsupervised domain adaptive target detection method based on center alignment and relationship significance |
CN112883714A (en) * | 2021-03-17 | 2021-06-01 | 广西师范大学 | ABSC task syntactic constraint method based on dependency graph convolution and transfer learning |
-
2019
- 2019-10-30 CN CN201911045261.4A patent/CN110880019B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3385887A1 (en) * | 2017-04-08 | 2018-10-10 | INTEL Corporation | Sub-graph in frequency domain and dynamic selection of convolution implementation on a gpu |
CN108062753A (en) * | 2017-12-29 | 2018-05-22 | 重庆理工大学 | The adaptive brain tumor semantic segmentation method in unsupervised domain based on depth confrontation study |
CN108288051A (en) * | 2018-02-14 | 2018-07-17 | 北京市商汤科技开发有限公司 | Pedestrian identification model training method and device, electronic equipment and storage medium again |
CA3002100A1 (en) * | 2018-04-18 | 2019-10-18 | Element Ai Inc. | Unsupervised domain adaptation with similarity learning for images |
US20190325299A1 (en) * | 2018-04-18 | 2019-10-24 | Element Ai Inc. | Unsupervised domain adaptation with similarity learning for images |
CN109753992A (en) * | 2018-12-10 | 2019-05-14 | 南京师范大学 | The unsupervised domain for generating confrontation network based on condition adapts to image classification method |
CN109948648A (en) * | 2019-01-31 | 2019-06-28 | 中山大学 | A kind of multiple target domain adaptive migration method and system based on member confrontation study |
CN109948741A (en) * | 2019-03-04 | 2019-06-28 | 北京邮电大学 | A kind of transfer learning method and device |
CN111259720A (en) * | 2019-10-30 | 2020-06-09 | 北京中科研究院 | Unsupervised pedestrian re-identification method based on self-supervision agent feature learning |
CN111222471A (en) * | 2020-01-09 | 2020-06-02 | 中国科学技术大学 | Zero sample training and related classification method based on self-supervision domain perception network |
CN111340021A (en) * | 2020-02-20 | 2020-06-26 | 中国科学技术大学 | Unsupervised domain adaptive target detection method based on center alignment and relationship significance |
CN112883714A (en) * | 2021-03-17 | 2021-06-01 | 广西师范大学 | ABSC task syntactic constraint method based on dependency graph convolution and transfer learning |
Non-Patent Citations (9)
Title |
---|
JUN LI等: "Utilizing GCN and Meta-Learning Strategy in Unsupervised Domain Adaptation for Pancreatic Cancer Segmentation", 《IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS》 * |
KAIYANG ZHOU等: "Learning Generalisable Omni-Scale Representations for Person Re-Identification", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
TIANZHU ZHANG: "GCAN: Graph Convolutional Adversarial Network for Unsupervised Domain Adaptation", 《2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
刘大鹏等: "结合源域差异性与目标域不确定性的深度迁移主动学习方法", 《模式识别与人工智能》 * |
周海波: "源域,目标域间语义网络距离对隐喻加工的影响", 《第十五届全国心理学学术会议论文摘要集》 * |
张天柱: "小规模数据集图像分类方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
彭雪莹等: "基于图卷积网络的迁移学习轴承服役故障诊断", 《计算机应用》 * |
徐晓义: "浅谈概念隐喻中源域和目标域的对应关系", 《现代交际》 * |
滕文秀: "面向高分辨率遥感影像场景深度特征的领域自适应方法研究", 《中国优秀博硕士学位论文全文数据库(硕士)农业科技辑》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111612051A (en) * | 2020-04-30 | 2020-09-01 | 杭州电子科技大学 | Weak supervision target detection method based on graph convolution neural network |
CN111612051B (en) * | 2020-04-30 | 2023-06-20 | 杭州电子科技大学 | Weak supervision target detection method based on graph convolution neural network |
CN112629863A (en) * | 2020-12-31 | 2021-04-09 | 苏州大学 | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions |
CN113011456B (en) * | 2021-02-05 | 2022-07-15 | 中国科学技术大学 | Unsupervised domain adaptation method based on class adaptive model for image classification |
CN113011456A (en) * | 2021-02-05 | 2021-06-22 | 中国科学技术大学 | Unsupervised domain adaptation method based on class adaptive model for image classification |
CN113128667A (en) * | 2021-04-02 | 2021-07-16 | 中国科学院计算技术研究所 | Cross-domain self-adaptive graph convolution balance migration learning method and system |
CN113128667B (en) * | 2021-04-02 | 2023-10-31 | 中国科学院计算技术研究所 | Cross-domain self-adaptive graph rolling balance migration learning method and system |
CN113436197A (en) * | 2021-06-07 | 2021-09-24 | 华东师范大学 | Domain-adaptive unsupervised image segmentation method based on generation of confrontation and class feature distribution |
CN113255823A (en) * | 2021-06-15 | 2021-08-13 | 中国人民解放军国防科技大学 | Unsupervised domain adaptation method and unsupervised domain adaptation device |
CN113255823B (en) * | 2021-06-15 | 2021-11-05 | 中国人民解放军国防科技大学 | Unsupervised domain adaptation method and unsupervised domain adaptation device |
US11797611B2 (en) | 2021-07-07 | 2023-10-24 | International Business Machines Corporation | Non-factoid question answering across tasks and domains |
CN113688867A (en) * | 2021-07-20 | 2021-11-23 | 广东工业大学 | Cross-domain image classification method |
CN113723917A (en) * | 2021-08-24 | 2021-11-30 | 中国人民解放军32382部队 | Association construction method and device of instrument management standard and instrument technical standard |
CN113723917B (en) * | 2021-08-24 | 2024-03-29 | 中国人民解放军32382部队 | Method and equipment for constructing association between instrument management standard and instrument technical standard |
CN113743523A (en) * | 2021-09-13 | 2021-12-03 | 西安建筑科技大学 | Visual multi-feature guided construction waste fine classification method |
CN114048546A (en) * | 2021-11-17 | 2022-02-15 | 大连理工大学 | Graph convolution network and unsupervised domain self-adaptive prediction method for residual service life of aircraft engine |
CN114048546B (en) * | 2021-11-17 | 2024-04-23 | 大连理工大学 | Method for predicting residual service life of aeroengine based on graph convolution network and unsupervised domain self-adaption |
Also Published As
Publication number | Publication date |
---|---|
CN110880019B (en) | 2022-07-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110880019B (en) | Method for adaptively training target domain classification model through unsupervised domain | |
CN110580501B (en) | Zero sample image classification method based on variational self-coding countermeasure network | |
CN110097094B (en) | Multiple semantic fusion few-sample classification method for character interaction | |
CN109741332A (en) | A kind of image segmentation and mask method of man-machine coordination | |
WO2019223384A1 (en) | Feature interpretation method and device for gbdt model | |
CN112766334B (en) | Cross-domain image classification method based on pseudo label domain adaptation | |
CN106951471B (en) | SVM-based label development trend prediction model construction method | |
CN103559504A (en) | Image target category identification method and device | |
CN109871504B (en) | Course recommendation system based on heterogeneous information network and deep learning | |
CN113140254B (en) | Meta-learning drug-target interaction prediction system and prediction method | |
CN111414461A (en) | Intelligent question-answering method and system fusing knowledge base and user modeling | |
CN104573130A (en) | Entity resolution method based on group calculation and entity resolution device based on group calculation | |
CN107004141A (en) | To the efficient mark of large sample group | |
CN110737805B (en) | Method and device for processing graph model data and terminal equipment | |
CN109919252A (en) | The method for generating classifier using a small number of mark images | |
CN114022904B (en) | Noise robust pedestrian re-identification method based on two stages | |
CN113656596A (en) | Multi-modal entity alignment method based on triple screening fusion | |
US20190026654A1 (en) | Bootstrapping multiple varieties of ground truth for a cognitive system | |
Hamim et al. | Student profile modeling using boosting algorithms | |
WO2020135054A1 (en) | Method, device and apparatus for video recommendation and storage medium | |
CN111339258B (en) | University computer basic exercise recommendation method based on knowledge graph | |
CN113239159A (en) | Cross-modal retrieval method of videos and texts based on relational inference network | |
Sameer et al. | Source camera identification model: Classifier learning, role of learning curves and their interpretation | |
CN111160077A (en) | Large-scale dynamic face clustering method | |
CN107577681A (en) | A kind of terrain analysis based on social media picture, recommend method and system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
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 |