Multi-view representation learning method and system based on adaptive view communication and multilevel residual connection learning

Abstract

The invention belongs to the technical field of artificial intelligence and machine learning, and discloses a multi-view representation learning method based on adaptive view communication and multilevel residual connection learning, and the method achieves the joint modeling of consistency and complementarity through sharing-specific information decomposition, simulates the human communication mechanism, and achieves the multi-view representation learning through multi-level residual connection learning. Multi-round communication is realized by a cross-view attention mechanism driven by specific information, and an adaptive stop condition is designed in combination with sharing similarity and mutual information to dynamically determine an optimal communication round number. Meanwhile, a multi-level residual fusion strategy is adopted, original features and interactive features are organically combined, and the sufficiency of representation and the calculation efficiency are both considered. According to the method, the performance of a multi-view learning task can be remarkably improved, and the method is suitable for multi-source data fusion scenes such as intelligent perception, medical diagnosis and automatic driving and has important scientific significance and engineering application value.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and machine learning technology, and in particular relates to a multi-view representation learning method and system based on adaptive view communication and multi-level residual connection learning. Background Technology

[0002] In practical applications, it is often necessary to process data from multiple sources simultaneously, such as LiDAR, cameras, and GPS in autonomous driving, and imaging data and genetic data in medical diagnosis. This type of data is called multi-view data. The goal of multi-view representation learning is to ensure that data from different perspectives is processed simultaneously. Figure 1 While ensuring consistency of information, we can effectively explore their complementarity to obtain a unified and highly discriminative feature representation.

[0003] There are three main types of existing methods:

[0004] Joint representation method: Directly splicing features from different views can take advantage of complementarity, but it ignores consistency and is prone to redundancy;

[0005] Alignment methods: These methods model consistency through correlation or distribution alignment, but often lose unique complementary information.

[0006] Sharing-specific approach: This approach breaks down information into shared and specific parts, balancing consistency and complementarity, but lacks cross-view communication mechanisms.

[0007] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0008] (1) It is difficult to make efficient use of consistency and complementarity at the same time.

[0009] (2) The communication and interaction mechanism is fixed and lacks adaptive adjustment, resulting in insufficient computing overhead or information utilization.

[0010] (3) There is a lack of dynamic interaction between views and a lack of information exchange between views. Summary of the Invention

[0011] To address the problems existing in the prior art, this invention provides a multi-view representation learning method based on adaptive view communication and multi-level residual connection learning.

[0012] This invention is implemented as follows: a multi-view representation learning method based on adaptive view communication and multi-level residual connection learning includes:

[0013] Step 1: Input multi-source, multi-view data;

[0014] Step 2: Model the features of each view as shared information and specific information;

[0015] Step 3: Construct a cross-view attention adaptive mechanism based on specific information to simulate human communication mechanisms and realize information interaction and dynamic communication between views with adaptive specific information weights.

[0016] Step 4: Dynamically control the number of communication rounds based on the similarity of shared information between the views before and after adaptive view communication and the mutual information of specific information, adaptively determine the stopping conditions, and evaluate the computational efficiency;

[0017] Step 5: A multi-level residual connection fusion strategy is adopted to combine the initial features and interaction features element by element, taking into account both semantic fidelity and information complementarity, to obtain the final unified representation.

[0018] Step 6: Input the unified representation into the classifier or predictor to complete the downstream task.

[0019] Furthermore, the shared information is aligned through similarity constraints to ensure consistency of different views in the semantic space.

[0020] Furthermore, the specific information is distinguished from shared information through orthogonal constraints to ensure the independence of complementary features.

[0021] Furthermore, the cross-view attention mechanism adaptively assigns weights based on the importance of specific information, simulating human communication mechanisms, thereby enabling dynamic information transfer, communication, and mutual assistance between views with adaptive weights for specific information.

[0022] Furthermore, the adaptive communication stopping condition consists of the following two parts:

[0023] (1) The cosine similarity between shared information reaches a preset threshold;

[0024] (2) The mutual information between different specific information reaches a preset threshold.

[0025] Furthermore, the multi-level residual fusion combines the original representation with the representation of the communication back view by adding elements one by one, so as to ensure semantic fidelity and generalization ability.

[0026] The proposed method is applicable to multi-view learning tasks such as classification, clustering, regression, and visualization, and can improve the completeness and discriminativeness of feature representations.

[0027] Another objective of this invention is to provide a multi-view representation learning system based on adaptive view communication and multi-level residual connection learning, comprising:

[0028] Multi-view data input module, used for inputting multi-view data;

[0029] The feature decomposition module is used to extract the shared representation and specific representation of multiple views separately using a non-linear learning network;

[0030] The adaptive view communication module is used for dynamic information interaction based on the specific representation of a view, employing a cross-view adaptive attention mechanism.

[0031] The communication control module is used to set a threshold based on the similarity of shared representations and mutual information of specific information, and to determine whether to continue communication.

[0032] The fusion representation module is used to gradually fuse initial features and interaction features through multi-level residual connections to obtain the final view feature representation;

[0033] The task prediction module is used to input fused representations into a classifier or other prediction model for practical tasks such as image and text classification and fault diagnosis.

[0034] Another object of the present invention is to provide a computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the steps of the multi-view representation learning method based on adaptive view communication and multi-level residual connection learning.

[0035] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the multi-view representation learning method based on adaptive view communication and multi-level residual connection learning.

[0036] Another objective of this invention is to provide an information data processing terminal for implementing the multi-view representation learning system based on adaptive view communication and multi-level residual connection learning.

[0037] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0038] 1. Shared-Specific Decomposition Mechanism: By adding strict loss function constraints, the data of each view is decomposed and extracted to obtain shared information (ensuring consistency) and specific information (providing complementarity).

[0039] 2. Information-Specific Attention Communication Mechanism: Utilizing specific information from each view as communication content, an attention mechanism is introduced to dynamically measure the importance of complementary information from different views, simulating information exchange in human communication, and achieving adaptive complementarity of view information.

[0040] 3. Adaptive view communication stopping condition: Based on the cosine similarity of the shared information of the views after multiple rounds of view communication and the mutual information of specific information, the number of communication rounds is dynamically determined to avoid feature redundancy and achieve a balance between computational efficiency and representation sufficiency.

[0041] 4. Multi-level residual fusion strategy: Through adaptive view communication rounds, multi-level residual connections are learned and element-wise added residual connections are used to mine deep-level features of the view. The original features and interaction features are fused, preserving the original semantics while improving interaction stability, thereby enhancing the model's generalization and robustness.

[0042] This method can significantly improve the performance of downstream tasks such as multi-view classification, clustering, and regression, and has broad application prospects in autonomous driving, medical assistance, fault diagnosis, and big data processing. Attached Figure Description

[0043] Figure 1 This is a flowchart of a multi-view representation learning method based on adaptive view communication and multi-level residual connection learning provided in an embodiment of the present invention.

[0044] Figure 2 This is a block diagram of a multi-view representation learning system based on adaptive view communication and multi-level residual connection learning provided in an embodiment of the present invention.

[0045] Figure 3 This is a schematic diagram of the downstream tasks of the system provided in an embodiment of the present invention.

[0046] Figure 4 This is a schematic diagram of a human communication model provided in an embodiment of the present invention.

[0047] Figure 5 This is a system adaptive view communication network structure diagram provided in an embodiment of the present invention.

[0048] Figure 6 This is a graph showing the change of the adaptive loss function trade-off parameters provided in an embodiment of the present invention.

[0049] Figures 7(a) and 7(b) are comparison diagrams of the technical effects obtained by the embodiments of the present invention.

[0050] Figures 8(a) and 8(b) show the classification results of the embodiment of the present invention on image and text multi-view datasets. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] like Figure 1 As shown, the multi-view representation learning method based on adaptive view communication and multi-level residual connection learning provided by this embodiment of the invention includes the following steps:

[0053] S101, Input multi-source, multi-view data;

[0054] S102, model the features of each view as shared information and specific information;

[0055] S103, based on specific information, constructs a cross-view attention adaptive mechanism to simulate human communication mechanism and realize information interaction and dynamic communication between views with adaptive specific information weights;

[0056] S104: Dynamically control the number of communication rounds based on the similarity of shared information between the preceding and following views and the mutual information of specific information in adaptive view communication, adaptively determine the stopping conditions, and evaluate computational efficiency;

[0057] S105 adopts a multi-level residual connection fusion strategy to combine the initial features and interaction features element by element, taking into account both semantic fidelity and information complementarity, to obtain the final unified representation.

[0058] S106 will uniformly represent the input classifier or predictor to complete downstream tasks.

[0059] The shared information provided in this embodiment of the invention is aligned through similarity constraints to ensure consistency of different views in the semantic space.

[0060] The specific information provided in this embodiment of the invention is distinguished from shared information through orthogonal constraints to ensure the independence of complementary features.

[0061] The cross-view attention mechanism provided in this invention adaptively assigns weights based on the importance of specific information, simulating human communication mechanisms, thereby enabling dynamic information transmission, communication, and mutual assistance between views with adaptive weights for specific information.

[0062] The adaptive communication stopping condition provided in this embodiment of the invention consists of the following two parts:

[0063] (1) The cosine similarity between shared information reaches a preset threshold;

[0064] (2) The mutual information between different specific information reaches a preset threshold.

[0065] The multi-level residual fusion provided in this embodiment of the invention combines the original representation with the representation of the communication post-view by adding elements one by one, so as to ensure semantic fidelity and generalization ability.

[0066] The proposed method is applicable to multi-view learning tasks such as classification, clustering, regression, and visualization, and can improve the completeness and discriminativeness of feature representations.

[0067] like Figure 2 As shown, an embodiment of the present invention provides a multi-view representation learning system based on adaptive view communication and multi-level residual connection learning, comprising:

[0068] Multi-view data input module, used for inputting multi-view data;

[0069] The feature decomposition module is used to extract the shared representation and specific representation of multiple views separately using a non-linear learning network;

[0070] The adaptive view communication module is used for dynamic information interaction based on the specific representation of a view, employing a cross-view adaptive attention mechanism.

[0071] The communication control module is used to set a threshold based on the similarity of shared representations and mutual information of specific information, and to determine whether to continue communication.

[0072] The fusion representation module is used to gradually fuse initial features and interaction features through multi-level residual connections to obtain the final view feature representation;

[0073] The task prediction module is used to input fused representations into a classifier or other prediction model for practical tasks such as image and text classification and fault diagnosis.

[0074] Another object of the present invention is to provide a computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the steps of the multi-view representation learning method based on adaptive view communication and multi-level residual connection learning.

[0075] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the multi-view representation learning method based on adaptive view communication and multi-level residual connection learning.

[0076] Another objective of this invention is to provide an information data processing terminal for implementing the multi-view representation learning system based on adaptive view communication and multi-level residual connection learning.

[0077] Specific implementation of the present invention:

[0078] The multi-view representation learning system based on adaptive view communication and multi-level residual connection learning mainly includes the following modules: Figure 2 )

[0079] 1. Multi-view data input module: Input multi-view data.

[0080] Feature decomposition module: Uses a non-linear learning network to extract shared and specific representations of multiple views respectively.

[0081] 2. Adaptive View Communication Module: Based on the specific representation of the view, it uses a cross-view adaptive attention mechanism for dynamic information interaction.

[0082] 3. Communication control module: Based on the similarity of shared representations and mutual information of specific information, a threshold is set to determine whether to continue communication.

[0083] 4. Fusion Representation Module: Through multi-level residual connections, the initial features and interactive features are gradually fused to obtain the final view feature representation.

[0084] 5. Task Prediction Module: This module integrates the input representations from classifiers or other prediction models for practical tasks such as image and text classification, and fault diagnosis.

[0085] Implementation steps

[0086] 1. Input Data Preparation: Input data from multiple views into the system, such as images, text, or sensor signals. Assume a multi-view image dataset X, where the... Each sample contains a view V represents the number of views. Different views represent data from different sources or perspectives, such as the color, texture, and edges of an image.

[0087] 2. Shared-Specific Information Decomposition: Based on the principles of consistency and complementarity of multi-view data, a representation learning network is used to extract initial features for each view, and similarity constraints are used to maintain the consistency of shared representations: consistency is represented as: Complementarity is expressed as: Where i,j represent the i-th and j-th elements of the multi-view dataset X. The first sample A view.

[0088] 3. Information-Specific Driven Communication: A cross-view attention mechanism is constructed, using specific information as the core and dynamically allocating weights to simulate human communication mechanisms, enabling information exchange and transmission between views, facilitating mutual assistance. This is based on the view... For example, the weighted complementary representation of other views is compared with the view. View communication is performed. The specific communication process is represented by the following formula:

[0089]

[0090]

[0091] in Indicates by view ,Right now The new view obtained after communication Indicates the splicing of features; It is to exclude the first The complementary information representations of views other than the first view are adaptively weighted using the corresponding attention weights and then concatenated to obtain the complementary information representation. Indicates the first A representation learning network is used to learn the representation of each view, thus completing one round of view communication.

[0092] 4. Adaptive Communication Stop: Calculate the cosine similarity of the current shared representation and the mutual information between different specific representations. When both reach a set threshold, communication stops. Assume the model learns a consistent representation for each view. and complementarity representation as follows:

[0093]

[0094]

[0095] In each round of communication, the representation from the previous round is used. and As input, the representation of the current round is updated via a communication mechanism. and complementarity representation :

[0096]

[0097] in The representation learning results of the corresponding view single-round communication Let be the number of samples. Between round t-1 and round t, a similarity comparison is performed between the consistent and complementary representations of the communication in round t-1 and round t to determine whether the optimal number of communication attempts has been reached. The similarity discriminant analysis calculation formula is as follows:

[0098]

[0099]

[0100] in, This represents the similarity measurement function. It sets the similarity threshold. When the t-wheel and When all reach the threshold, that is and Communication is interrupted at this time. A pre-defined convergence threshold indicates when communication is considered to have ended once the similarity reaches a certain level. After communication stops, the final view representation is output. ,in This is the final round.

[0101] 5. Residual Fusion: This involves fusing the original representation with the interaction representation obtained after multiple rounds of communication element-wise, preserving the original semantics and improving robustness. The final residual fusion is expressed by the following formula:

[0102]

[0103] in Indicates by view ,Right now The new view obtained after K rounds of adaptive view communication is the final residual fusion representation. This represents the concatenation of features, where K represents the final communication round. The shared information in the K-round communication view is merged level by level and element by element.

[0104] 6. Task Prediction: The fused representation is input into the classifier to complete downstream tasks such as classification, regression, or clustering. Figure 3 )

[0105] Modeling and optimization details

[0106] 1. A View Communication Model Inspired by Human Communication Mechanisms: Based on human communication mechanisms, the human communication model is represented as ( Figure 4 ):

[0107] Each communicator, represented as a view, simultaneously receives or sends dialogue.

[0108] The communicators share some common background and common knowledge, which in Figure 4 The term "in Chinese" represents shared knowledge.

[0109] Each communicator has their own knowledge base or area of ​​experience. Figure 4 This represents his way of thinking.

[0110] Effective communication is achieved through multiple rounds of information exchange. All ideas are dynamic during these exchanges. Figure 3 The term "interaction" in this context refers to a multi-turn dialogue between the communicators.

[0111] 2. Network architecture of adaptive view communication model: ( Figure 5 )

[0112] First, two sub-networks are built for each view: one for modeling shared information and the other for modeling specific information.

[0113] Shared information is aligned through similarity constraints to ensure consistency among multiple views; specific information is distinguished from shared information through orthogonal constraints to ensure the independence of complementary features.

[0114] Specific information from each view is stored in a specific information pool, and complementary information is adaptively obtained from other views through an attention mechanism;

[0115] By using residual fusion, the acquired weighted complementary information is fused with the original features of the view element by element to obtain a new representation of the view features, thus completing one round of adaptive view communication.

[0116] Multi-turn adaptive view communication overlay, similar to multi-turn dialogue in human communication mechanisms, mines deep view features and obtains efficient view feature representations through full communication.

[0117] 3. Loss Function Design and Joint Optimization:

[0118] Classification loss: Cross-entropy loss is used to supervise the classification of the final fused representation.

[0119]

[0120] in It's a real label. It is the probability distribution predicted by the model, and n is the number of classification categories.

[0121] Similarity loss: Using adversarial similarity constraints to maximize the similarity of shared information between different views and reduce cross-view differences.

[0122]

[0123] in Let Z represent the similarity function, and Z be the consistency representation of the view.

[0124] Orthogonal loss: Using orthogonal constraints, the independence of shared information and specific information is strictly constrained, separating shared information from specific information and preventing information redundancy.

[0125]

[0126] Where ||·|| represents the Frobenius norm, H is the complementary representation of the view, and Z is the consistent representation of the view.

[0127] The overall loss function is the weighted sum of the above three terms:

[0128]

[0129] in These represent classification loss, similarity loss, and orthogonality loss, respectively. α and β are adaptive dynamic trade-off parameters for the loss function, with their weights changing adaptively according to different training stages. (See appendix for details) Figure 6 )

[0130] By using an end-to-end training approach and employing the backpropagation algorithm to jointly optimize network parameters, efficient feature representations are provided for downstream tasks.

[0131] 4. Model Training and Application:

[0132] The final merged multi-view representation is then input into the classifier;

[0133] The model is optimized using the aforementioned joint loss function during the training phase;

[0134] During the inference phase, only adaptive communication and residual fusion need to be performed to obtain an efficient and unified feature representation for tasks such as classification, clustering, decision-making, and regression.

[0135] To verify the effectiveness of the multi-view collaborative contrastive representation learning method (AVCMRL) of this invention, performance evaluations were conducted on several typical multi-source datasets. The experimental data included the Food Image Recognition Dataset (FOOD), the Natural Image Dataset (CIFAR), the Multimodal Object Recognition Dataset (NUSOBJ), and the Text Topic Classification Dataset (REUTERS), covering different modal feature scenarios such as images and text, fully demonstrating the universality and stability of the algorithm in cross-modal and multi-view data processing.

[0136] As shown in Figure 7(a), on the FOOD multi-view dataset, the method of this invention achieves the best results in both classification accuracy and F1-score, with an accuracy of 74.38% and an F1-score of 0.738. Compared with existing methods such as MVLDAN, MVDN, DLDA, Co-Em, SIMM, and MRL-VC, this represents an average improvement of approximately 5–10 percentage points. These results demonstrate the significant advantages of this invention in feature alignment and modality consistency constraints.

[0137] As shown in Figure 7(b), the results on the CIFAR multi-view dataset also demonstrate the superiority of the algorithm. AVCMRL achieves a classification accuracy of 73.24% and an F1-score of 0.732 on this dataset, representing improvements of approximately 9.3% and 0.1 respectively compared to the highest level of existing algorithms. This fully demonstrates that the method of this invention can maintain stable cross-domain generalization performance when facing image features with different distributions.

[0138] Figures 8(a) and 8(b) show the t-SNE visualization results of the model of this invention and the traditional method in the feature embedding space. The comparison reveals that the sample distribution extracted by the traditional method shows significant overlap and blurred inter-class boundaries, while the feature distribution generated by the algorithm of this invention exhibits clear inter-class separation and tight intra-class aggregation. This demonstrates that AVCMRL effectively improves feature discriminativeness through the multi-view contrast constraint mechanism.

[0139] Validation results on downstream tasks using the NUSOBJ and REUTERS multi-view datasets demonstrate that the proposed technique not only performs excellently in image classification but also achieves high accuracy in text and multimodal joint classification tasks. Cluster visualization (Figure 8(b)) further shows the model's consistency enhancement effect in the multi-view feature space, validating the robustness and transferability of the method in cross-modal tasks.

[0140] The AVCMRL algorithm proposed in this invention significantly improves the multimodal feature fusion effect by introducing collaborative contrastive learning and modality invariant constraints in multi-view scenarios. Compared with previous methods, it achieves significant improvements in classification accuracy, F1-score, and feature discriminancy, providing high-precision and scalable algorithmic support for applications such as face recognition, autonomous driving, multi-source sensor fusion, and medical auxiliary diagnosis.

[0141] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0142] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-view representation learning method based on adaptive view communication and multi-level residual connection learning, characterized in that, The method comprises the following steps: Step 1, inputting multi-source multi-view data; Step 2, modeling each view feature as shared information and specific information; Step 3, constructing a cross-view attention adaptive mechanism based on specific information, simulating human communication mechanism, and realizing information interaction and dynamic communication between views with adaptive specific information weight; Step 4, dynamically controlling the number of communication rounds according to the similarity of shared information and the mutual information of specific information before and after adaptive view communication, adaptively determining the stopping condition, and evaluating the calculation efficiency; Step 5, adopting a multi-level residual connection fusion strategy to combine the initial feature and the interaction feature element by element, giving consideration to semantic fidelity and information complementarity, and obtaining a final unified representation; Step 6, inputting the unified representation into a classifier or a predictor to complete a downstream task. 2.The multi-view representation learning method based on adaptive view communication and multi-level residual connection learning of claim 1, wherein, The shared information is aligned by similarity constraint to ensure the consistency of different views in the semantic space. 3.The method of claim 1, wherein, The specific information is distinguished from the shared information by orthogonal constraint to ensure the independence of complementary features. 4.The method of claim 1, wherein, The cross-view attention mechanism adaptively allocates weights according to the importance of specific information, simulates human communication mechanism, and realizes dynamic information transmission, communication and mutual assistance between views with adaptive specific information weight. 5.The method of learning multi-view representation based on adaptive view communication and multi-level residual connection learning according to claim 1, wherein, The adaptive communication stopping condition consists of the following two parts: (1) the cosine similarity between shared information reaches a preset threshold; (2) the mutual information between different specific information reaches a preset threshold. 6.The method of learning multi-view representation based on adaptive view communication and multi-level residual connection learning according to claim 1, wherein, The multi-level residual fusion combines the original representation and the representation of the view after communication by element-wise addition to ensure semantic fidelity and generalization ability. The method is suitable for multi-view learning tasks such as classification, clustering, regression and visualization, and can improve the integrity and discriminability of feature representation.

7. An adaptive view-communication and multi-level residual-connection learning based multi-view representation learning system implementing the adaptive view-communication and multi-level residual-connection learning based multi-view representation learning method according to any one of claims 1-6. The multi-view representation learning system based on adaptive view communication and multi-level residual connection learning comprises: a multi-view data input module for inputting multi-view data; a feature decomposition module for extracting shared representation and specific representation of multi-view by using a nonlinear learning network; an adaptive view communication module for dynamic information interaction by using a cross-view adaptive attention mechanism based on the specific representation of the view; a communication control module for setting a threshold according to the similarity of shared representation and the mutual information of specific information to determine whether to continue communication; a fusion representation module for gradually fusing the initial feature and the interaction feature by multi-level residual connection to obtain the final view feature representation; a task prediction module for inputting the fusion representation into a classifier or other prediction model for actual tasks such as image and text classification and fault diagnosis.

8. A computer device, comprising: The computer device comprises a memory and a processor, and the memory stores a computer program, which, when executed by the processor, causes the processor to execute the steps of the multi-view representation learning method based on adaptive view communication and multi-level residual connection learning according to any one of claims 1-6. 9.A computer readable storage medium storing a computer program, the computer program, when executed by a processor, causing the processor to perform the steps of the method of claim 1-6.

10. An information data processing terminal, characterized by The information data processing terminal is configured to implement the system of claim 7.

Images