Reliable classification method based on deep fuzzy multi-view learning

Abstract

The invention discloses a reliable classification method based on deep fuzzy multi-view learning, and the method specifically comprises the following steps: collecting multi-view data, carrying out the preprocessing of the data, and obtaining a processed training data set and a test data set; constructing a deep fuzzy multi-view learning neural network model; training the neural network model to obtain a trained neural network model; and predicting the category of each sample in the test data set through the trained neural network model, and outputting the uncertainty of a corresponding classification result. According to the method, through a deep fuzzy learning mechanism, the classification accuracy of conflict views is remarkably improved, meanwhile, the uncertainty of classification results is output, and reliable confidence reference is provided for users; the innovative deep fuzzy multi-view learning method not only improves the performance of a multi-view classification technology, but also provides a new thought and direction for the development of related technologies in the future, and has important academic value and wide application prospects.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-view data classification, and in particular relates to a reliable classification method based on deep fuzzy multi-view learning. Background Art

[0002] In today's complex and ever-changing data processing and classification tasks, multi-view learning technology has become a research hotspot due to its ability to integrate multi-source information from different data sources or sensors. Multi-view data typically includes a combination of images, text, audio, and other formats, describing the same object or scene from different perspectives. The core goal of multi-view learning is to leverage the complementarity of these multi-source data to improve the performance of classification, prediction, or other machine learning tasks. However, in practical applications, existing multi-view learning methods for multi-view classification face challenges with data conflicts and insufficient uncertainty estimation. On the one hand, data from different views may conflict due to acquisition errors, noise, or incomplete data sources, making it difficult for the model to accurately determine the sample category, resulting in reduced classification performance. On the other hand, in safety-critical scenarios such as medical diagnosis and autonomous driving, traditional methods are unable to effectively estimate the uncertainty of classification results. Simply providing classification results cannot meet user needs, limiting their application. Therefore, reliable multi-view classification has important research significance and practical application value for the development of artificial intelligence in safety-critical fields.

[0003] In practical applications, traditional multi-view classification models face numerous challenges. First, data between different views may conflict or be inconsistent. For example, some views may provide contradictory information due to data acquisition errors, noise, or incomplete data sources. The presence of these conflicting views makes it difficult for traditional multi-view classification models to accurately determine the true class of a sample, resulting in a significant decline in classification performance. Second, existing multi-view classification methods often lack effective estimates of the uncertainty in the classification results for instances with conflicting views. In many safety-critical real-world scenarios, users also need to understand the model's confidence in the classification results so that further verification or intervention measures can be taken when necessary. Although current AI-based multi-view classification algorithms have achieved significant progress in classification performance, most of these methods assume that data is perfectly aligned. However, real-world data often contains errors and faults, especially in multi-view scenarios, where data quality varies greatly across views. This imperfection in the data makes traditional multi-view classification models perform poorly when dealing with conflicting views, making it difficult to accurately estimate the uncertainty of the classification results, thus failing to meet the high reliability and confidence requirements of safety-critical scenarios. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a reliable classification method based on deep fuzzy multi-view learning, which solves the problem that existing multi-view classification methods are difficult to accurately classify conflicting multi-view instances and cannot provide accurate uncertainty estimation results, thus making it difficult to apply in actual scenarios.

[0005] In order to solve the above technical problems, the present invention is implemented in the following ways:

[0006] A reliable classification method based on deep fuzzy multi-view learning includes the following steps:

[0007] S1. Collect multi-view data, pre-process the data, and obtain the processed training data set and test data set;

[0008] S2, build a deep fuzzy multi-view learning neural network model;

[0009] S3. Input a clean and aligned multi-view training dataset to train the neural network model to obtain a trained neural network model.

[0010] S4. Predict the category of each sample in the test data set containing the conflicting views through the trained neural network model, and output the uncertainty of the corresponding classification results.

[0011] Furthermore, the specific method of step S1 is as follows:

[0012] In multi-view classification, the original data For N samples and V views, the min-max method is used to normalize the data and adjust the value range of all values ​​to the real number interval [0,1]. The expression is as follows:

[0013]

[0014] in, represents the normalization function, represents the v-th view data value of the n-th sample;

[0015] The preprocessed data is then divided into a training set and a test set, with the ratio of the training set to the test set being 7:3.

[0016] Furthermore, the specific method of step S2 is as follows:

[0017] For each view v, build its view-specific deep fuzzy neural network model , and use a randomly initialized fully connected layer network as the classifier , extract the membership of the view to different categories , where k represents the kth category, and the specific membership expression is as follows:

[0018]

[0019] Among them, ReLU represents the rectified linear unit activation function, and its expression is as follows:

[0020]

[0021] Membership degree to each view The multi-view fusion function is used for aggregation, and its expression is as follows:

[0022]

[0023] in, represents the uncertainty of the classification result of the v-th view of the n-th sample, represents the conflict degree of the classification result of the v-th view of the n-th sample;

[0024] Uncertainty The expression is as follows:

[0025]

[0026] in, It represents the category credibility of the v-th view of the n-th sample for the k-th category. Its calculation formula during neural network training is as follows:

[0027]

[0028] in, Represents the category label of the v-th view of the n-th sample for the k-th category. If it belongs to the k-th category, it is 1, otherwise it is 0;

[0029] When testing the neural network The calculation formula is as follows:

[0030]

[0031] The expression of the conflict degree between the classification result of the vth view of the nth sample and the classification results of other views is as follows:

[0032]

[0033] Furthermore, the specific method of step S3 is as follows:

[0034] Input clean and aligned multi-view data into the neural network, train the neural network, and perform a specific neural network for each view. and its classifier , the loss function expression is as follows:

[0035]

[0036] in, Indicates the amount of data for neural network training once;

[0037] For the entire neural network, the overall loss function expression is as follows:

[0038]

[0039] Adopt Adam optimizer, according to the overall loss of training Optimize model parameters and complete the training of the deep model to obtain the trained model ; The calculation formula of the optimized deep fuzzy neural network parameters is as follows:

[0040]

[0041] in, Represents the parameters of the deep fuzzy neural network, AdamOptimizer represents the Adam reverse gradient optimizer, Represents the learning rate.

[0042] Furthermore, the specific method of step S4 is as follows:

[0043] Multi-view data with noise and misaligned views , input the trained neural network , the neural network outputs the classification result and the corresponding uncertainty , the calculation formula is as follows:

[0044] .

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] Through a deep fuzzy learning mechanism, the present invention significantly improves the classification accuracy of conflicting view instances and can simultaneously output the uncertainty of the classification results, providing users with a reliable confidence reference. This feature makes it more practical and reliable in safety-critical real-world scenarios (such as medical diagnosis, autonomous driving, financial risk assessment, etc.), and can effectively deal with problems such as data collection errors, noise interference, or incomplete data sources, demonstrating strong robustness.

[0047] The innovative deep fuzzy multi-view learning method not only improves the performance of multi-view classification technology, but also provides new ideas and directions for the development of related technologies in the future. It has important academic value and broad application prospects. It is applicable to a variety of multi-view data types, including combinations of images, text, audio, etc., and can be widely used in multiple fields such as medicine, autonomous driving, and finance. At the same time, it expands the application of fuzzy learning in the multimedia field, opening up a new application space for deep fuzzy learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the process of the classification method of the present invention. DETAILED DESCRIPTION

[0049] The specific implementation of the present invention will be further described in detail below with reference to the accompanying drawings and specific examples.

[0050] like Figure 1 As shown in FIG, a reliable classification method based on deep fuzzy multi-view learning includes the following steps:

[0051] S1. Collect multi-view data and pre-process the data to obtain the processed training dataset and test dataset. The specific method is as follows:

[0052] In multi-view classification, the original data For data with N samples and V views, in order to eliminate the influence of different dimensions and orders of magnitude on the analysis results, the min-max method is used to normalize the data and adjust the value range of all values ​​to the real number interval [0,1]. The expression is as follows:

[0053]

[0054] in, represents the normalization function, represents the v-th view data value of the n-th sample;

[0055] The preprocessed data is then divided into a training set and a test set, with the ratio of the training set to the test set being 7:3.

[0056] S2. Build a deep fuzzy multi-view learning neural network model. The specific method is as follows:

[0057] For each view v, build its view-specific deep fuzzy neural network model , and use a randomly initialized fully connected layer network as the classifier , extract the membership of the view to different categories , where k represents the kth category, and the specific membership expression is as follows:

[0058]

[0059] Among them, ReLU represents the rectified linear unit activation function, and its expression is as follows:

[0060]

[0061] Where x represents a real number;

[0062] In order to aggregate the classification results of each view, a dual-reliable multi-view fusion function is used to calculate the membership of each view. The multi-view fusion function is used for aggregation, and its expression is as follows:

[0063]

[0064] in, represents the uncertainty of the classification result of the v-th view of the n-th sample, represents the conflict degree of the classification result of the v-th view of the n-th sample;

[0065] Uncertainty The expression is as follows:

[0066]

[0067] in, It represents the category credibility of the v-th view of the n-th sample for the k-th category. Its calculation formula during neural network training is as follows:

[0068]

[0069] in, Represents the category label of the v-th view of the n-th sample for the k-th category. If it belongs to the k-th category, it is 1, otherwise it is 0;

[0070] When testing the neural network The calculation formula is as follows:

[0071]

[0072] The expression of the conflict degree between the classification result of the vth view of the nth sample and the classification results of other views is as follows:

[0073]

[0074] S3. Input a clean and aligned multi-view training dataset and train the neural network model to obtain a trained neural network model. The specific method is as follows:

[0075] Input clean and aligned multi-view data into the neural network, train the neural network, and perform a specific neural network for each view. and its classifier , the loss function expression is as follows:

[0076]

[0077] in, Indicates the amount of data for neural network training once;

[0078] For the entire neural network , the overall loss function expression is as follows:

[0079]

[0080] Adopt Adam optimizer, according to the overall loss of training Optimize model parameters and complete the training of the deep model to obtain the trained model ; The calculation formula of the optimized deep fuzzy neural network parameters is as follows:

[0081]

[0082] in, Represents the parameters of the deep fuzzy neural network, AdamOptimizer represents the Adam reverse gradient optimizer, Represents the learning rate.

[0083] S4. Use the trained neural network model to predict the category of each sample in the test dataset containing the conflicting views and output the uncertainty of the corresponding classification results. The specific method is as follows:

[0084] Multi-view data with noise and misaligned views , input the trained neural network , the neural network outputs the classification result and the corresponding uncertainty , the calculation formula is as follows:

[0085] .

[0086] The above description is merely an embodiment of the present invention. It is stated again that, for a person skilled in the art, several improvements can be made to the present invention without departing from the principles of the present invention, and these improvements are also included in the scope of protection of the claims of the present invention.

Claims

1. A reliable classification method based on deep fuzzy multi-view learning, characterized by: The specific steps include: S1. Collect multi-view data, pre-process the data, and obtain the processed training data set and test data set; S2, build a deep fuzzy multi-view learning neural network model; S3. Input a clean and aligned multi-view training dataset to train the neural network model to obtain a trained neural network model. S4. Predict the category of each sample in the test data set containing the conflicting views through the trained neural network model, and output the uncertainty of the corresponding classification results.

2. The reliable classification method based on deep fuzzy multi-view learning according to claim 1, characterized in that: The specific method of step S1 is as follows: In multi-view classification, the original data For N samples and V views, the min-max method is used to normalize the data and adjust the value range of all values ​​to the real number interval [0,1]. The expression is as follows: in, represents the normalization function, represents the v-th view data value of the n-th sample; The preprocessed data is then divided into a training set and a test set, with the ratio of the training set to the test set being 7:

3.

3. The reliable classification method based on deep fuzzy multi-view learning according to claim 1, characterized in that: The specific method of step S2 is as follows: For each view v, build its view-specific deep fuzzy neural network model , and use a randomly initialized fully connected layer network as the classifier , extract the membership of the view to different categories , where k represents the kth category, and the specific membership expression is as follows: Among them, ReLU represents the rectified linear unit activation function, and its expression is as follows: Membership degree to each view The multi-view fusion function is used for aggregation, and its expression is as follows: in, represents the uncertainty of the classification result of the v-th view of the n-th sample, represents the conflict degree of the classification result of the v-th view of the n-th sample; Uncertainty The expression is as follows: in, It represents the category credibility of the v-th view of the n-th sample for the k-th category. Its calculation formula during neural network training is as follows: in, Represents the category label of the v-th view of the n-th sample for the k-th category. If it belongs to the k-th category, it is 1, otherwise it is 0; When testing the neural network The calculation formula is as follows: The expression of the conflict degree between the classification result of the vth view of the nth sample and the classification results of other views is as follows: 。 4. The reliable classification method based on deep fuzzy multi-view learning according to claim 1, characterized in that: The specific method of step S3 is as follows: Input clean and aligned multi-view data into the neural network, train the neural network, and perform a specific neural network for each view. and its classifier , the loss function expression is as follows: in, Indicates the amount of data for neural network training once; For the entire neural network, the overall loss function expression is as follows: Adopt Adam optimizer, according to the overall loss of training Optimize model parameters and complete the training of the deep model to obtain the trained model ; The calculation formula of the optimized deep fuzzy neural network parameters is as follows: in, Represents the parameters of the deep fuzzy neural network, AdamOptimizer represents the Adam reverse gradient optimizer, Represents the learning rate.

5. The reliable classification method based on deep fuzzy multi-view learning according to claim 1, characterized in that: The specific method of step S4 is as follows: Multi-view data with noise and misaligned views , input the trained neural network , the neural network outputs the classification result and the corresponding uncertainty , the calculation formula is as follows: 。

Images