Data privacy protection method and system

Abstract

The present invention discloses a data privacy protection method and system. The system includes a dual variational autoencoder, a coupled generative adversarial network, and a discriminator. The dual variational autoencoder is used to encode raw data of different types and different feature dimensions into features of a latent space with the same feature dimensions; the coupled generative adversarial network is used to generate latent space features corresponding to the raw data; the coupled generative adversarial network includes a generator for discrete data and a generator for continuous data, as well as a time series fully connected layer and a fully connected layer. The high-order features of the latent space output by the generator for discrete data and the generator for continuous data are input into the time series connected layer, the output of the time series fully connected layer is input into the fully connected layer, and the output of the fully connected layer is input into a comparator; the output of the comparator is decoded and input into the discriminator to output a label for distinguishing synthetic data from real data.

Description

Technical Field

[0001] The present invention relates to a data privacy protection method for medical data and also to a corresponding data privacy protection system, belonging to the technical field of medical care informatics. Background Art

[0002] Data synthesis technology generates synthetic data by simulating real-world data. It protects privacy while preserving the data's utility. This technology is particularly important in the medical field. It can generate synthetic data based on statistical data from medical imaging, protecting patient privacy while facilitating data sharing. Furthermore, this synthetic data can be used to train algorithms, thereby improving the efficiency of medical services.

[0003] Data synthesis techniques rely on a variety of models, including variational autoencoders, generative adversarial networks, and diffusion models, aiming to improve the availability of synthetic data. Numerous models have been proposed and demonstrated effectiveness in synthesizing both continuous data (such as vital signs) and discrete data (such as medical interventions). However, the quality of medical data synthesis is challenging due to the complex relationships between these two types of data, making it difficult for existing methods to simultaneously capture their distributions and their interrelationships.

[0004] While the primary advantage of synthetic data is preventing personal privacy leaks, synthetic methods based on raw data still pose privacy risks when facing attacks such as reverse engineering and re-identification. The privacy protection issues faced by synthetic data are similar to those faced by machine learning models, including data theft, attribute inference, and membership inference attacks. Deep generative models are generally vulnerable to these attacks, while adopting privacy protection measures such as differential privacy sacrifices model accuracy, resulting in a decrease in the quality and usability of synthetic data compared to raw data. Therefore, improving the quality and usability of synthetic data while protecting privacy is a major challenge facing current synthetic data technologies. Summary of the Invention

[0005] The primary technical problem to be solved by the present invention is to provide a data privacy protection system for medical data.

[0006] Another technical problem to be solved by the present invention is to provide a data privacy protection method for medical data.

[0007] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0008] According to a first aspect of an embodiment of the present invention, a data privacy protection system is provided, comprising a dual variational autoencoder, a coupled generative adversarial network, and a discriminator; wherein,

[0009] The dual variational autoencoder is used to encode raw data of different types and different feature dimensions in the medical field into features of a latent space with the same feature dimension; the raw data includes discrete data and continuous data in a medical scenario, wherein the continuous data is patient vital sign data, and the discrete data is binary identification data of medical intervention measures; the dual variational autoencoder includes an encoder for continuous data, a decoder for continuous data, an encoder for discrete data, a decoder for discrete data, a continuous data sampling module located between the encoder and decoder for continuous data, a discrete data sampling module located between the encoder and decoder for discrete data, and a comparator;

[0010] The coupled generative adversarial network is used to generate latent space features corresponding to the original data; wherein the coupled generative adversarial network includes a generator for discrete data and a generator for continuous data, as well as a time series fully connected layer and a fully connected layer, the high-order features of the latent space output by the generator for discrete data and the generator for continuous data are input to the time series connected layer, the output of the time series fully connected layer is input to the fully connected layer, and the output of the fully connected layer is input to the comparator;

[0011] The discriminator includes a continuous data discriminator and a discrete data discriminator. The output of the comparator is decoded and input into the discriminator to output a label for distinguishing synthetic data from real data.

[0012] Preferably, the coupled generative adversarial network includes a BLSTM network for continuous data and a BLSTM network for discrete data that are coupled to each other, and also includes a fully connected layer for continuous data and a fully connected layer for discrete data.

[0013] The BLSTM network of continuous data is used as a generator of continuous data, and continuous random noise is input to obtain its The hidden layer output at the moment is input to the fully connected layer of continuous data;

[0014] The BLSTM network of discrete data is used as a generator of discrete data and is input with discrete random noise to obtain its The hidden layer output at the moment is input to the fully connected layer of the discrete data;

[0015] exist At time t, the inputs of the BLSTM network for continuous data and the BLSTM network for discrete data both include: The hidden layer output of the BLSTM network for continuous data at time t, and The hidden layer output of the BLSTM network of discrete data at the moment is obtained The hidden layer output at time t.

[0016] Preferably, the inputs of the continuous data encoder and the discrete data encoder are respectively The real data at the time, and the time The corresponding decoder outputs the reconstructed data and hidden state tensor when ;

[0017] The outputs of the continuous data encoder and the discrete data encoder are sampled respectively to obtain the time The latent space high-order features are input into the corresponding decoders respectively;

[0018] The decoders are based on time The hidden state tensor of The latent space high-order features, decoding generation moment The reconstruction data,

[0019] The comparator is based on the time Real data and time The reconstructed data is used for comparative learning.

[0020] Preferably, the tensor of continuous data output by the time series fully connected layer and the tensor of discrete data are input into the fully connected layer; the fully connected layer outputs the final latent space high-order feature representation.

[0021] Preferably, the patient's vital signs data include at least one of heart rate, blood pressure, blood oxygen saturation, and respiratory rate; and the medical intervention measures include binary status identifiers of ventilator use, antihypertensive drug use, and pressor drug use.

[0022] According to a second aspect of an embodiment of the present invention, a data privacy protection method is provided, wherein continuous random noise and historical vital sign data are input to a generator of continuous data to obtain The hidden layer output at time ; Input discrete random noise and medical intervention status at historical moments into the discrete data generator to obtain The hidden layer output at time ;

[0023] Will Input the continuous fully connected layer to get the continuous real data The corresponding high-order features of the latent space ;Will Input discrete fully connected layer to get discrete real data The corresponding high-order features of the latent space ;

[0024] The high-order features of the latent space and Input to the time series fully connected layer, The output at each moment is a tensor group , split to get the tensor of continuous data and discrete data tensors ,in Indicates from Until the moment The time step of the moment;

[0025] Will 、 Input to the fully connected layer to output the final latent space high-order feature representation and to the comparator;

[0026] The encoder for continuous data and the encoder for discrete data complete the original data respectively. The encoding is sampled by the continuous data sampling module and the discrete data sampling module respectively, and the moment with the same feature dimension is obtained The latent space high-order features are decoded to obtain the reconstructed data, which is input to the comparator;

[0027] After comparative learning by the comparator, they are encoded separately and input into the continuous data discriminator and the discrete data discriminator to output labels for distinguishing synthetic data from real data.

[0028] Compared with the existing technology, the present invention combines dual variational autoencoders, coupled generative adversarial networks and discriminators to generate synthetic data that is highly similar to the original data while protecting personal privacy. It is particularly suitable for processing continuous and discrete data in the medical field. The system not only improves the timing and accuracy of synthetic data, but also further enhances the quality of synthetic data through adversarial training and discriminator evaluation. In addition, the present invention also introduces a differential privacy protection mechanism and a dynamic privacy budget allocation strategy to balance privacy protection and model performance, ensuring that the quality of synthetic data does not decrease as training progresses. Experimental results confirm that the present invention can generate synthetic data with similar distribution, strong correlation and high availability while protecting privacy, showing significant technical effects and practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A schematic diagram of the overall architecture of the CoGAN model provided in the first embodiment of the present invention;

[0030] Figure 2 for Figure 1 Schematic diagram of the architecture of the dual variational autoencoder in ;

[0031] Figure 3 for Figure 1 Schematic diagram of the coupled recurrent network in [1].

[0032] Figure 4 for Figure 1 Schematic diagram of the architecture of the coupled generative adversarial network in

[15] ;

[0033] Figure 5 Schematic diagram of the architecture of the continuous data discriminator;

[0034] Figure 6 This is a schematic diagram of the structure of the DP-CoGAN model that implements dynamic privacy budget allocation in an embodiment of the present invention;

[0035] Figure 7A This is the Pearson correlation coefficient effect diagram of 24 hours of real data;

[0036] Figure 7B For Figure 7A The corresponding Pearson correlation coefficient effect diagram of the synthetic data;

[0037] Figure 8 The privacy protection effects of the three models under different privacy budgets. DETAILED DESCRIPTION

[0038] The technical content of the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] For the convenience of describing the technical solution of the present invention, the following description assumes Medical data of each patient For the patients, whose medical data are represented as ,in, represents the length of the time series, From 1 o'clock to All medical data of the patient at the time, including continuous data such as blood pressure and heart rate monitored by instruments ; It also includes discrete data indicating whether there are medical intervention measures, which takes the value of 1 (intervention) or 0 (no intervention), that is, Specifically, for ,patient The continuous data is represented as ,in Represents the dimension of continuous features, It means that the patient At the moment The collected Similarly, the patient The discrete data is represented as ,in Represents the dimension of continuous features, It means that the patient At the moment The collected The value of the dimension data.

[0040] First embodiment

[0041] Medical data can be mainly divided into two categories: (1) continuous data describing patients’ vital signs and laboratory test data; (2) discrete data indicating whether medical intervention was adopted.

[0042] Table 1 schematically shows several common types of medical data.

[0043] Table 1 Classification of medical data types

[0044]

[0045] It's not difficult to find correlations between different data types. For example, the use of antihypertensive medications can affect changes in blood pressure, and the blood pressure value at a certain moment can also affect the use of antihypertensive medications. Therefore, when completing the task of generating medical data, it is not possible to train two types of models completely independently and complete the generation task independently. However, due to the different formats of continuous data and discrete data, the two feature tensors cannot be simply spliced ​​together to train the model. In addition, a patient's vital signs will not change suddenly. At a certain moment, the patient's vital signs are affected by the vital signs at the previous moment on the one hand, and by whether or not medical intervention has occurred on the other hand. Therefore, the influence of these two factors should be considered simultaneously when training the generation model.

[0046] like Figure 1 As shown, the data privacy protection system (abbreviated as CoGAN model) provided by the first embodiment of the present invention includes a dual variational autoencoder ( Figure 1 The dotted line in the upper middle) and the coupled generative adversarial network ( Figure 1 The part outside the dotted line) and the discriminator. Among them, the dual variational autoencoder is used to encode the original data of different types and different feature dimensions in the medical field into features of the latent space with the same feature dimension; the original data includes discrete data and continuous data in the medical scenario, wherein the continuous data is the patient's vital signs data, and the discrete data is the binary identification data of the medical intervention measures. The patient's vital signs data here include at least one of heart rate, blood pressure, blood oxygen saturation, and respiratory rate; the medical intervention measures include binary state identification of ventilator use, antihypertensive drug use, and pressor drug use. The coupled generative adversarial network is used to generate latent space features corresponding to the original data. In Figure 1In the figure, the variational autoencoder and coupled generative adversarial network (GAN) within the dotted box on the left are used to generate continuous data; the variational autoencoder and coupled GAN within the dotted box on the right are used to generate discrete data. Training the dual variational autoencoder uses contrastive learning, while training the coupled GAN combines features from different data types, improving the similarity between the synthesized data and the original data while preserving the correlation between continuous and discrete data.

[0047] The Dual Variational Autoencoder (DVAE) is used to encode continuous data (such as vital signs) and discrete data (such as medical interventions) in medical data into the same latent space, addressing the heterogeneity of data types. By jointly encoding the shared latent space, it can capture complex relationships between the two types of data (such as the causal relationship between blood pressure changes and drug interventions), overcoming the limitations of existing technologies that process different types of data independently.

[0048] Specifically, the dual variational autoencoder transforms continuous data and discrete data Encoded as high-order features of the same latent space . and The task of generating models is to generate raw data Transformed into high-order features of the generated latent space , thereby solving the problem of inconsistent data types. In addition, when training the dual variational autoencoder, in addition to considering the variational lower bound loss, feature matching loss and contrastive learning loss are also considered to strengthen the feature The connection between them.

[0049] like Figure 2 As shown, the dual variational autoencoder in the embodiment of the present invention includes: an encoder for continuous data , decoder for continuous data , encoder for discrete data , a decoder for discrete data , a continuous data sampling module located between the continuous data encoder and decoder, a discrete data sampling module located between the discrete data encoder and decoder, and a comparator. The inputs of the two encoders are the time Real data at the time (continuous real data) or (discrete real data), and the moment The corresponding decoder outputs the reconstructed data and hidden state tensor when ; the two encoders (continuous encoders and discrete encoders ) output, and after sampling, we can get the time The latent space high-order features ( and ), respectively input to the corresponding decoder (continuous decoder and discrete decoders ). Each decoder is based on the time The hidden state tensor and time The latent space high-order features, decoding generation moment Reconstruction data (continuous reconstruction data and discrete reconstruction data ). Based on the time Real data and time The reconstructed data of the comparator is compared with the learning loss to measure the similarity and difference between the real data and the reconstructed data, thus achieving the goal of contrastive learning.

[0050] and The original data will be completed separately The encoding is sampled by the continuous data sampling module and the discrete data sampling module respectively to obtain the encoding data with the same feature dimension. . Then the decoder Complete the encoding data to produce an output similar to the original data , The comparator measures the similarity between the reconstructed data and the positive samples, and the difference between the reconstructed data and the negative samples, allowing the model to distinguish between different categories of data and extract information that can represent the characteristics of the data.

[0051] Figure 2 in Represents a continuous loss, which is an unsupervised generative model, so the label is the sample itself. It is calculated from the continuous sample itself and the reconstructed sample. Similarly, the discrete loss Calculated from the discrete samples themselves and the reconstructed samples. This figure shows that after pre-training, the encoded results are fed into the GAN for training. Here, Z is the latent space representation, and X is the original data space representation. The GAN learns the latent space representation. Contrastive learning is used in pre-training, using end-to-end reconstructions of the original data space to calculate the loss.

[0052] As can be seen, the decoder transforms high-order features in the latent space back into the original data space, learning effective representations of the data and reconstructing the data from these representations. The decoder explores the latent space and generates new data samples. This helps learn the underlying features of the data, which can be used for downstream tasks. Furthermore, the decoder output can be used to evaluate model performance. Using a comparator to compare the original and reconstructed data, we can extract information that represents the data characteristics and quantify the model's ability to capture these features.

[0053] Specifically, the moment Encoder for continuous data The input consists of three parts: 1) time Real data at the time , 2) Moment Decoder for continuous data Output (reconstructed data) ,3) Encoder for continuous data Moment The hidden state tensor of Continuous data encoder The output is the corresponding time Latent space high-order feature representation That is, at the moment When the continuous data encoder The output can be expressed as follows.

[0054]

[0055] in, The input is the real data of the corresponding time In addition, it also includes the decoder output of the previous moment , sig represents the Sigmoid function. Decoder for time-continuous data The input includes: time Latent space high-order feature representation ,time The hidden state tensor of , so the decoder for continuous data The output can be expressed as follows.

[0056]

[0057] Similarly, for discrete data, the encoder of discrete data The output of can also be expressed as formula 1; the decoder of discrete data The output can also be expressed as Equation 2.

[0058] The loss function of the dual variational autoencoder is in addition to the variational lower bound loss In addition, it also includes feature matching loss and contrastive learning loss . Variational lower bound loss It includes the sum of reconstruction loss and KL divergence, where the reconstruction loss is usually calculated using mean squared error (MSE), while the KL divergence measures the difference between the latent variable distribution output by the encoder and the standard normal distribution. The loss functions include cross entropy loss (CE), L1 distance or mean absolute error (MAE). Contrastive learning loss functions can use triple loss, InfoNCE loss, etc. In the following description, examples are given to illustrate 、 、 However, this does not constitute a limitation to the present invention, and other loss functions may also be used according to actual needs.

[0059] In one embodiment of the present invention, feature matching loss is used To measure the distance of features in the latent space. Specifically, the encoder and The sample Encoded as a set of feature tensors in the latent space .because Indicates the characteristics of the same patient, which are semantically similar, so The distance in the latent space should also be closer. Therefore, in the embodiment of the present invention, the Euclidean distance between the two tensors is used as the feature matching loss to measure the The distance between them.

[0060]

[0061] In one embodiment of the present invention, the two decoders output to the comparator for contrastive learning. Contrastive learning loss (Contra) is used In this embodiment of the present invention, NT-Xent loss (Normalized Temperature-scaled Cross Entropy Loss) is used, but other losses can also be used.

[0062] Here, the features from the same sample in the latent space are Defined as a pair of positive samples, and features from different samples Specifically, assuming there are N samples, after the dual variational autoencoder, N sets of feature tensors in the latent space can be obtained. Since the feature dimensions of the latent space features of continuous data and discrete data are the same, all the features of the latent space can be regarded as a set , .use and Represents different types of features (continuous or discrete), that is .in, Represents the characteristics of continuous data Or the characteristics of discrete data ,in ; accordingly, Can be expressed with Corresponding or , In particular, you can use Represents the characteristics of any continuous data and the characteristics of discrete data , So, for patients In terms of is a positive sample pair, and the rest The group sample pair is a negative sample pair, and its contrastive learning loss function is:

[0063] .

[0064] in, express cosine similarity between ; is the temperature normalization factor, ; indicator vector , if and only if The value is 1. The overall contrast loss function can be expressed as:

[0065]

[0066] In summary, the loss function of the dual variational autoencoder can be expressed as:

[0067]

[0068] in is the weight of the corresponding loss function term, represents the variational lower bound loss, represents the feature matching loss, Denotes contrastive learning loss. The encoder for continuous data can capture the causal relationship between blood pressure changes and drug intervention through variational lower bound loss and contrastive learning loss.

[0069] Dual learning is achieved through the dual variational autoencoder described above. In the dual variational autoencoder, two variational autoencoders process continuous and discrete data respectively. This allows the model to learn the interrelationships and transformations between the two types of data. This also helps the model better complete and predict incomplete or missing data, thereby improving the model's generalization ability and generating complete and accurate medical data. This capability is particularly important in medical data analysis and prediction, helping doctors and researchers better understand and utilize complex medical data.

[0070] Next, we will further explain the coupled generative adversarial network in the CoGAN model.

[0071] As mentioned above, when processing medical data, the temporal characteristics of the medical data should be fully considered, and the correlation between continuous data and discrete data should also be considered. Specifically, the embodiment of the present invention improves the accuracy of generated data by the following methods:

[0072] (1) Medical data includes continuous data describing the patient’s vital signs and discrete data describing whether medical intervention has occurred. The floating point number between The original data is obtained by the aforementioned dual variational autoencoder The latent space representation of The coupled generative adversarial network is used as a generator, whose generation goal is to generate original data Generate corresponding The latent space features (2) Medical data has strong temporal correlation. The value at a certain moment is often affected by the previous moment. In order to make the generative model learn this temporal correlation and improve the quality of synthetic data, a temporal fully connected layer is added to fully consider the temporal correlation. Data generated at all times Received Time data impact.

[0073] The difficulty in generating medical data is continuous data With discrete data Interrelated. Tensor group It can be encoded into a high-order feature representation in the same latent space through a dual variational autoencoder Then the generator (coupled generative adversarial network) receives a random noise to generate a high-order feature representation tensor group in the latent space .

[0074] In generating high-order feature representation tensor groups When we consider the tensor With tensors In one embodiment of the present invention, a coupled recurrent neural network (CRN) is used as part of a generator to learn the associations between different data. Two RNN networks are coupled together to construct a CRN network, which supports learning features from two sets of data in a cyclic manner.

[0075] Since ordinary LSTM networks can learn long-term dependency information but cannot learn the features of two types of data at the same time, the embodiments of the present invention optimize the bidirectional long short-term memory network (BLSTM network for short). By adding hidden states, the BLSTM network can learn the features of different tensor groups and use the BLSTM network to construct a coupled recurrent network.

[0076] Compared with the LSTM network, the BLSTM network in the embodiment of the present invention adds a new hidden state ( or ) input and pass the weight matrix , when the update of the gate control unit is completed.

[0077] At the moment , continuous data can be expressed as:

[0078]

[0079] Discrete data can be represented as:

[0080] ,

[0081] in and are two sets of BLSTM networks coupled to each other, and It is used to generate continuous data. Used to generate discrete data. Figure 3 As shown, and The hidden layer output and It serves not only as the input of the same group of BLSTM networks at the next moment, but also as the input of another group of BLSTM networks.

[0082] Specifically, combined Figure 1 and Figure 3 As shown, the BLSTM network for continuous data ( Figure 3 mid-dotted line) at time Receive random noise , the last moment network The hidden state , the network of discrete data at the previous moment ( Figure 3 The hidden state (shown by the dotted line) . Indicates the cell state at the current time step. BLSTM network for continuous data At the moment The state update formula is as follows.

[0083]

[0084] in, Indicates the input gate at time step The activation value of Indicates the forget gate at time step The activation value of Indicates the output gate at time step The activation value of Represents the candidate cell state at time step Candidate values ​​for Represents the cell state at time step The value of represents the hidden state at time step The value of Indicates the current time step The input vector of the LSTM network of continuous data (random noise); Represents the time step The hidden state value of the BLSTM network for discrete data; and Represents the weight matrix, different superscripts and subscripts represent different gates and different input sources; represents the bias vector; Represents the sigmoid activation function, used to compress values ​​between 0 and 1; tanh represents the hyperbolic tangent activation function, used to compress values ​​between -1 and 1; ⊙ represents element-wise multiplication (Hadamard product), used for gating mechanisms. Except for the above, all other BLSTM networks are continuous data. The value of .

[0085] Similarly, for the BLSTM network of discrete data , at the moment Receive random noise , the last moment network The hidden state , BLSTM network of continuous data at the previous moment The hidden state . BLSTM network for discrete data At the moment The state update formula is as follows:

[0086]

[0087] in, Indicates that the input gate is at time step The activation value of Indicates that the forget gate is at time step The activation value of Indicates that the output gate is at time step The activation value of Represents the cell state at time step Candidate values ​​for Represents the cell state at time step The value of represents the hidden state at time step The value of Indicates the current time step The input vector of the LSTM network of continuous data (random noise); Represents the time step The hidden state value of the BLSTM network for continuous data; and Represents the weight matrix, different superscripts and subscripts represent different gates and different input sources; represents the bias vector; Represents the sigmoid activation function, which is used to compress values ​​between 0 and 1; tanh represents the hyperbolic tangent activation function; ⊙ represents element-wise multiplication.

[0088] It can be seen that the BLSTM network in the embodiment of the present invention includes a set of LSTM networks for continuous data and a set of LSTM networks for discrete data, respectively generating two types of data. Among them, in the LSTM network for continuous data provided by the embodiment of the present invention, the activation values ​​of the forget gate, input gate and output gate are all obtained by using two hidden states. and and random noise The resulting model can use the hidden states of the LSTM network for continuous data and the hidden states of the LSTM network for discrete data to correlate the two types of data. It can also calculate bidirectional hidden states by running the LSTM layers forward and backward along the time axis. In each set of BLSTM networks, both past and future information are processed simultaneously, thereby better capturing the contextual relationships in sequential data. Therefore, this BLSTM network, through bidirectional hidden state transfer, integrates the temporal correlations between continuous and discrete data (for example, how a previous moment's vital signs influence current intervention decisions), generating data that better aligns with real-world medical logic.

[0089] like Figure 4 As shown, the generator architecture based on the aforementioned BLSTM network includes a generator for continuous data and discrete data generators .

[0090] Generator for continuous data BLSTM network including continuous data and fully connected layers for continuous data .exist BLSTM network for continuous data at time The input includes: continuous random noise 、 time The hidden layer output as well as BLSTM network for discrete time data The hidden layer output , get its The hidden layer output at time .Will Input continuous fully connected layer ,get Generator for continuous time data Output , that is, continuous real data The corresponding high-order features of the latent space (See the formula below).

[0091]

[0092] Generator for discrete data It consists of two components: BLSTM network for discrete data and fully connected layers for discrete data .exist At time t, the BLSTM network of discrete data The input includes: discrete random noise 、 time The hidden layer output as well as time The hidden layer output , get its The hidden layer output at time .Will Input discrete fully connected layer ,get Moment Generator Output (See the formula below).

[0093]

[0094] Generator and High-order features of the output latent space and , is input into the temporal fully connected layer .exist Moment, Generator and generators Output tensor array , Until the moment Moment Generator and generators Output tensor array . With tensors Concatenate to get tensor Input , exist The output at each moment is a tensor group , split to get the tensor of continuous data and discrete data tensors . Among them, the dimensions of all tensor groups are the same.

[0095] The output y of a temporal fully connected layer is similar to that of a conventional fully connected layer, satisfying the following equation: y = Wx + b, where Wx is a matrix multiplication, representing the dot product of the weight matrix W and the input vector x; b is a bias vector, added to the output of the weight matrix to adjust the output offset. However, unlike a conventional fully connected layer, the input vector is two tensors of the same dimension. After being concatenated and fed into the fully connected layer, the output vector is split into two output vectors of equal length.

[0096] Output of the temporal fully connected layer 、 is input into the fully connected layer . exist Tensor that receives continuous data at all times and discrete data tensors As input, and output the final latent space high-order feature representation and , see the following formula:

[0097]

[0098] Through the above measures, the temporal fully connected layer of the coupled generative adversarial network can dynamically associate the vital signs data at the current moment with the medical intervention status at the previous moment.

[0099] For continuous data and discrete data, train continuous data discriminators separately and discrete data discriminator The output of the comparator is decoded by the decoder and input to the discriminator.

[0100] like Figure 5 As shown, the discriminator It consists of two parts: a recurrent neural network composed of LSTM and a fully connected layer. The input is real data Or generate data , the output is the label , so the discriminator can be expressed as follows :

[0101] .

[0102] Discriminator Structure and discriminator Similar, I will not go into details here.

[0103] It's important to note that the discriminator's goal is to output labels. These labels are the result of comparing and distinguishing generated data (synthetic data) from real data. Through comparative learning, the discriminator assesses whether the generated data is sufficiently realistic—that is, whether it can closely mimic the characteristics and distribution of real data. If the discriminator can accurately output labels that distinguish between real and synthetic data, this indicates that the synthetic data has successfully preserved the key characteristics of the original data while not leaking information sufficient to identify the original data. Therefore, the ability to output labels is a key metric for measuring the effectiveness of a data privacy protection system.

[0104] In addition, the generator , generator High-order feature representation of the latent space of each output 、 , cannot be used directly as a discriminator 、 Instead of input, it is first input to the decoder 、 , and obtain the final generated data 、 , and then input to the corresponding discriminator 、 .

[0105] The loss function of the generator is:

[0106]

[0107] in, represents the loss function of the generator, Represents the distribution Random noise sampled in The expected value obtained by averaging is and represent continuous or discrete random noise, respectively. and represent the generator of continuous or discrete data respectively, and represent continuous or discrete data discriminators, respectively.

[0108] Loss function of the discriminator Satisfy (where ):

[0109] During the training phase, the generator and discriminator are updated alternately. The generator continuously learns how to generate more realistic samples, while the discriminator continuously learns how to better distinguish between real samples and generated samples. This adversarial process drives improvements in the quality of samples generated by the generator.

[0110] Second embodiment

[0111] The second embodiment of the present invention provides a data privacy protection method that utilizes a dual variational autoencoder, a fully connected layer, and generators and discriminators for the two data types to improve the correlation between continuous and discrete data. Furthermore, a temporal connection layer is used to improve the temporal nature of the generated data, and random noise is added to the generator to enhance privacy.

[0112] Specifically, the data privacy protection method includes the following steps:

[0113] The continuous data generator is fed with continuous random noise and vital signs characteristics of historical moments, specifically continuous random noise 、 time The hidden layer output as well as BLSTM network for discrete time data The hidden layer output , get The hidden layer output at time ; Input discrete random noise and medical intervention status characteristics at historical moments into the generator of discrete data, specifically discrete random noise 、 time The hidden layer output as well as time The hidden layer output , get The hidden layer output at time Here, the temporal correlation between vital sign characteristics and medical intervention status characteristics is jointly modeled by coupling a bidirectional long short-term memory network, making it easier to generate synthetic data that conforms to medical logic.

[0114] Will Input the continuous fully connected layer to get the continuous real data The corresponding high-order features of the latent space ;Will Input discrete fully connected layer to get discrete real data The corresponding high-order features of the latent space ;

[0115] The high-order features of the latent space and Input to the time series fully connected layer, The output at each moment is a tensor group , split to get the tensor of continuous data and discrete data tensors ,in Indicates from Until the moment The time step of the moment;

[0116] Will 、 Input to the fully connected layer to output the final latent space high-order feature representation and to the comparator;

[0117] Use the encoder for continuous data and the encoder for discrete data to complete the original data respectively. The encoding is sampled by the continuous data sampling module and the discrete data sampling module to obtain the moment with the same feature dimension The latent space high-order features are decoded to obtain the reconstructed data, which is input to the comparator;

[0118] After comparative learning by the comparator, they are encoded separately and input into the continuous data discriminator and the discrete data discriminator to output labels for distinguishing synthetic data from real data.

[0119] The implementation process of the data privacy protection method includes a pre-training phase and a training phase. In the pre-training phase, the dual variational autoencoder is first trained, and the loss function is the loss function of the dual variational autoencoder. Then, in the training phase, for the identification of real data, the real data is first trained. Encoding to obtain high-order feature representation of the latent space , then Input decoder, get , and respectively Input Discriminator ; In adversarial training, through the generator Generating high-level feature representations ,Will Input decoder , , get generated data , and generate data respectively Input Discriminator , and obtain the identification result.

[0120] Based on the CoGAN model provided in the first embodiment of the present invention, the present invention further proposes an improved DP-CoGAN model. Figure 6 As shown in Figure 2, the DP-CoGAN model adds a differential privacy protection mechanism to the CoGAN model. This is achieved by adding random noise during the model training phase, thereby protecting the privacy of individuals in the training data. Compared to the CoGAN model, the DP-CoGAN model uses differential privacy to train the generative adversarial network, ensuring that the data generated by the generative network also meets the differential privacy requirements.

[0121] Specifically, given the DP-CoGAN model training set and training process satisfy − Differential privacy, so Generates synthetic data for differential privacy generation models It is also differentially private. Intuitively, a differentially private generative model should not directly access raw sensitive data during training, while also generating samples that are usable for specific downstream tasks. Currently, the mainstream solution is to use a differentially private stochastic gradient descent algorithm to complete model training.

[0122] In the DP-CoGAN model, the degree of data privacy protection is determined by the privacy budget Decide, The smaller the value, the higher the degree of data privacy protection. When performing the differential privacy stochastic gradient descent algorithm, each round of iteration needs to be based on the noise scale. Determines the size of the Gaussian noise added, and Determined by the privacy budget allocated to that iteration. As the number of algorithm iterations increases, the total amount of noise added to the DP-CoGAN model increases, making the model more difficult to converge. Therefore, it is necessary to dynamically adjust the amount of noise added to avoid degradation of the synthesized data quality as training rounds increase. This embodiment of the present invention leverages the multiple iterations of deep learning networks to balance the privacy budget allocated to each round and reduce algorithmic overhead, thus proposing a dynamic privacy budget allocation strategy.

[0123] To avoid adding excessive noise to the model as the number of iterations increases, the embodiment of the present invention gradually increases the privacy budget during the training phase and reduces the noise scale of each iteration. Below, two noise attenuation algorithms are proposed:

[0124] , or, .

[0125] in, represents the initial noise size, represents the expected noise size at the end of training, represents the training round, Indicates the decay rate.

[0126] In the above noise attenuation algorithm, the noise As the training rounds As the noise decreases, the privacy budget will gradually increase, but the noise cannot be reduced indefinitely. In exponential noise decay, the noise will not be less than ; In the power function noise attenuation, the parameter Control the attenuation range, when the number of training rounds When the noise Decay to ;when When the noise is maintained constant.

[0127] In addition, the initial noise level , attenuation rate Parameters such as need to be determined at the beginning of training. For data with high privacy protection strength, a larger initial noise and a smaller decay rate should be selected; similarly, for applications with high requirements for synthetic data quality, a smaller initial noise and a larger decay rate should be selected.

[0128] In one embodiment of the present invention, an adaptive clipping threshold selection algorithm (ACTS algorithm) based on Gaussian mechanism is further proposed. In this algorithm, the samples are divided into multiple batches for training. First, let is equal to the maximum gradient norm in the current batch of samples, that is, Then, according to the input parameters , the algorithm will be the interval Divide equally, obtain subintervals; again, for each interval , calculate the number of samples of the gradient norm in this interval, and use Indicates that ; Finally, for any interval and ,Towards Add obedience Gaussian noise distribution, and the gradient clipping threshold is determined as the maximum frequency of the interval .

[0129] The ACTS algorithm can solve abnormal samples The impact on the overall clipping threshold can also make it comply with the differential privacy specification. On the one hand, abnormal samples Only the frequencies of two intervals can be changed at most. On the other hand, by adding Gaussian noise , differential privacy protection can be achieved. At the same time, due to the clipping threshold determined by the ACTS algorithm , so it can be considered that the optimal threshold is selected, making the algorithm converge more easily. Therefore, the clipping threshold calculates the average gradient size of all samples after processing. The threshold is neither too small, resulting in excessive clipping and information loss, nor too large, resulting in insufficient privacy protection. Therefore, it can better balance privacy protection and model performance.

[0130] Next, the technical effects of the data privacy protection method provided by the embodiment of the present invention are further introduced through experimental data.

[0131] The validation experiments of this invention used the MIMIC-III dataset, constructed by the Computational Physiology Laboratory at the Massachusetts Institute of Technology. The MIMIC-III dataset contains data on ICU patients at Beth Israel Deaconess Medical Center between 2001 and 2012. The dataset primarily includes demographic data, vital sign measurements, laboratory test results, treatment options, length of hospital stay, and other information.

[0132] The data preprocessing was implemented based on the preprocessing framework proposed by the MIMIC-EXTRACT program. As shown in Table 2, the MIMIC-EXTRACT program can extract four tables from the raw data:

[0133] Table 2 Dataset table structure

[0134]

[0135] After obtaining the data in the four tables, data processing is required to extract features, handle outliers, and normalize the data to obtain data suitable for model training. The data screening conditions include:

[0136] (1) Patient age ≥16 years

[0137] (2) 24 hours ≤ patient stay ≤ 240 hours

[0138] Entering the raw data into the MIMIC-EXTRACT program generates four tables. The patients table is indexed by patient ID (subj_id), hospitalization ID (hadm_id), and ICU ID (icustay_id), while the remaining three tables are indexed by subj_id, hadm_id, icustay_id, and hours_in. Based on the data screening criteria, this embodiment of the present invention generates 14 discrete features and 114 continuous features. Because not every patient has all features, it is necessary to handle outliers such as null values ​​and select the last 24 hours of a patient's records. This embodiment then aggregates the continuous and discrete features by hour. Finally, the data is normalized and the maximum and minimum values ​​of each feature are saved to complete the reverse mapping of the generated data.

[0139] After the above processing, the data to be analyzed that can be used for model training obtained in the embodiment of the present invention is shown in Table 3.

[0140] Table 3 Statistics of data preprocessing results

[0141]

[0142] To accurately evaluate the quality of synthesized data, the embodiments of the present invention mainly conduct experiments around three aspects: the similarity in distribution between synthesized data and original data, the correlation between continuous data (such as vital signs data) and discrete data (such as medical intervention measures), and the usability of synthesized data in practical applications.

[0143] First, regarding data authenticity, that is, the similarity in distribution between the synthesized data and the original data, this embodiment of the present invention not only compares the overall differences between the two types of data but also provides a more detailed and intuitive comparison based on specific indicators. In comparing overall distribution differences, this embodiment of the present invention uses Maximum Mean Discrepancy (MMD) and a discriminative score to evaluate the performance of continuous and discrete data, respectively. The MMD primarily assesses the differences in distribution between the synthesized and original data, while the discriminative score evaluates the authenticity of the synthesized data based on the difficulty of distinguishing between the synthesized and original data.

[0144] MMD is often used to measure the consistency of two distributions. It measures the distance between two distributions in Hilbert space. The smaller the maximum mean difference, the smaller the distance between the two distributions and the closer the distributions are; the larger the maximum mean difference, the larger the distance between the two distributions. The calculation process can be expressed as follows:

[0145]

[0146] in, and Represents two probability distributions. The kernel function expressed as follows can embed the probability distribution into different reproducing Hilbert spaces.

[0147]

[0148] The core idea behind discriminant scoring is to train a binary discriminator separately from the generative model, using a portion of the training data as a test set. After the generative model is trained, the binary discriminator is trained using synthetic data (labeled 0) and test data (labeled 1). The binary discriminator is used to discriminate against the synthetic data. The accuracy of the binary discriminator is the discriminant score. Therefore, the optimal discriminant score is 0.5.

[0149] Secondly, this embodiment of the present invention uses the Pearson pairwise correlation coefficient (PCC) to assess the correlation between vital sign data and medical intervention data. The Pearson correlation coefficient is widely used to measure the degree of correlation between two variables. The value of the Pearson correlation coefficient ranges from -1 to 1. The closer the PCC is to 1, the more positive the correlation between the features; the closer it is to -1, the more negative the correlation between the features.

[0150] Again, the embodiment of the present invention trains a set of machine learning models on real data and generated data, including logistic regression (LR), decision tree (DT), random forest (RF) and multi-layer perceptron (MLP) models. It also calculates the true positive (TP), false positive (FP), true negative (TN) and false negative (FN) of each model, and calculates the sensitivity of the model accordingly. , specificity , positive predictive value and negative predictive value In terms of indicator selection, the AUC score and PR-AUC score were calculated by the ROC curve and Precision-Recall curve respectively as model evaluation indicators.

[0151] In terms of the selection of comparison models, since most models cannot generate both continuous and discrete data, for continuous data, this embodiment of the present invention selects three models for comparison: R(C)GAN, TimeGAN, and DP-CGAN. These are all typical GAN ​​models for generating time series data, and DP-CGAN is also a model for generating medical data. For discrete data, the comparison models selected are medGAN, seqGAN, and DP-CGAN, which perform better in generating discrete data.

[0152] In the design of the ablation experiment, the embodiment of the present invention mainly verifies the impact of the coupled recurrent network based on the BLSTM network on the generated data. Therefore, the ablation experiment uses the following model:

[0153] The VAE-GAN model consists of a dual variational autoencoder and a generative adversarial network. The dual variational autoencoder is the same as described above, while the generator in the generative adversarial network uses a standard LSTM network, which only considers the characteristics of a single class of data.

[0154] CoGAN model: The complete model proposed in this embodiment of the present invention uses a BLSTM network to construct a coupled recurrent network as a generator compared to the VAE-GAN model.

[0155] The experimental results are as follows:

[0156] (1) Authenticity evaluation of continuous synthetic data

[0157] As shown in the table, the MMD scores of the continuous synthetic data generated by the CoGAN model proposed in this embodiment of the present invention are compared with the MMD values ​​of synthetic data generated by existing algorithms such as R(C)GAN, Time-GAN, and DP-CGAN.

[0158] Table 4 Comparison of MMD values ​​of continuous synthetic data

[0159]

[0160] Comparative experiments show that the CoGAN model proposed in this embodiment of the present invention outperforms other models in generating continuous time series data. Ablation experiments show that the generator based on coupled recurrent networks performs better when faced with generating data of different categories.

[0161] As shown in Table 5, the embodiment of the present invention also uses a discrimination score to evaluate the generator. Among them, the discrimination score value of the CoGAN model is 0.755, which is better than the other three models and also better than the VAE-GAN model.

[0162] Table 5 Discrimination scores of continuous synthetic data

[0163]

[0164] (2) Authenticity Assessment of Discrete Synthetic Data

[0165] When evaluating discrete synthetic data, since discrete data takes values ​​of 0 or 1, the maximum mean difference cannot be used. Therefore, this embodiment of the present invention uses discriminant scoring to evaluate discrete values. As shown in Table 6, the CoGAN model proposed in this embodiment of the present invention also outperforms models such as DP-CGAN on discrete synthetic data. Ablation experiments also show that data generated using a coupled recurrent network outperforms data generated using a standard LSTM.

[0166] Table 6 Discriminant scores of discrete data

[0167]

[0168] The embodiment of the present invention also calculates the mean, variance, and KS-Statts of each distribution, which represents the difference between distributions, to reflect the performance of the synthetic data on specific indicators, as shown in Table 7.

[0169] Table 7 Comparison of mean and standard deviation of real data and generated data

[0170]

[0171] KS-Stats stands for the Kolmogorov-Smirnov statistic, a statistic used to compare the difference between two distributions. As can be seen, in the comparative experiments with R(C)GAN, Time-GAN, and DP-CGAN, the CoGAN model performed best across all six metrics.

[0172] (3) Relevance assessment

[0173] In order to evaluate whether there is a correlation between continuous data and discrete data in the synthetic data, the embodiment of the present invention carried out a correlation experiment. Specifically, the embodiment of the present invention uses the Pearson correlation coefficient (PCC) to evaluate the correlation. The embodiment of the present invention selects 24 hours of real data ( Figure 7A ) and synthetic data ( Figure 7B ) to calculate the Pearson correlation coefficient. As can be seen in the figure, there is a negative correlation between diastolic blood pressure and vasopressor medication in both the real and synthetic data. This is because a decrease in diastolic blood pressure leads to medical intervention; that is, the lower the diastolic blood pressure, the higher the likelihood of vasopressor use. Furthermore, a positive correlation is observed between heart rate and respiratory rate. This is because an increase in heart rate increases oxygen consumption, which is regulated by the nervous system, and respiratory rate also increases accordingly.

[0174] (4) Usability evaluation

[0175] Table 8 shows the usability evaluation results using original data, data generated by CoGAN, data generated by DP-CGANS, data generated by TimeGAN, and data generated by R(C)GAN. Logistic regression (LR), decision tree (DT), random forest (RF), and multilayer perceptron (MLP) models were trained and the Area Under Current Analysis (AUC) and PR-AUC scores were calculated. In tests on the Adult, Census, and Mimic-III datasets, the CoGAN model achieved higher scores than the other generative models.

[0176] Table 8 Detailed results of usability evaluation

[0177]

[0178] Because there is still correlation between synthetic data and real data, there may be some potential privacy leakage risks, including membership inference attacks, re-identification attacks, and attribute inference attacks. To address membership inference attacks, the embodiment of the present invention divides the original data into a training set (50%) and a test set (50%), and then trains the model using only the training set. After generating the synthetic data, a k-nearest neighbor (kNN) model is trained on the synthetic data. It is assumed that the attacker has all the real data, including training data and test data. Using the kNN model, the nearest neighbor is identified for each sample in the real dataset. A minimum threshold (for example, the minimum Hamming distance) is used to predict whether the real data sample belongs to the training data. The experimental results are shown in Table 9. As can be seen, regardless of the distance metric used, the average difference between the DP-CoGAN model provided by the embodiment of the present invention and the optimal score does not exceed 5%.

[0179] Table 9 Privacy attack experiment results

[0180]

[0181] In addition, the embodiment of the present invention also studies the impact of different privacy budgets on privacy protection. Given a model and a record, when the attacker knows that the record is used for model training, it is considered as information leakage. The embodiment of the present invention selects the CoGAN model without differential privacy, the DP-CoGAN model with differential privacy, and the DP-CGAN model in the prior art for experimentation. When implementing the member inference attack, a model with the same structure as the discriminator is used as a shadow model, and then a fully connected network is constructed as the attack model to evaluate the privacy protection effects of the three models under different privacy budgets. The results are as follows: Figure 8 As shown. Figure 8 In the figure, the horizontal axis represents the privacy budget , the vertical axis represents the accuracy of membership inference attack, where the closer it is to 0.5, the better the model is at preventing membership inference attack and the higher the degree of protection for individual privacy. Figure 8 It can be seen that even when When adding differential privacy, the accuracy of the DP-CoGAN model never exceeded 0.6, while the accuracy of the CoGAN model without differential privacy reached above 0.7, which shows that the privacy protection effect is obvious.

[0182] It should be noted that the above embodiments are merely examples, and the technical solutions of the various embodiments may be combined, and the order of the steps may be changed, all within the scope of protection of the present invention.

[0183] The above describes in detail the data privacy protection method and system provided by the present invention. For those skilled in the art, any obvious modification made to the present invention without departing from the essence of the present invention will constitute an infringement of the patent rights of the present invention and will result in corresponding legal liability.

Claims

1. A data privacy protection system, characterized in that It includes dual variational autoencoder, coupled generative adversarial network and discriminator; among them, The dual variational autoencoder is used to encode raw data of different types and different feature dimensions in the medical field into features of a latent space with the same feature dimension; the raw data includes discrete data and continuous data in a medical scenario, wherein the continuous data is patient vital sign data, and the discrete data is binary identification data of medical intervention measures; the dual variational autoencoder includes an encoder for continuous data, a decoder for continuous data, an encoder for discrete data, a decoder for discrete data, a continuous data sampling module located between the encoder and decoder for continuous data, a discrete data sampling module located between the encoder and decoder for discrete data, and a comparator; The coupled generative adversarial network is used to generate latent space features corresponding to the original data; wherein the coupled generative adversarial network includes a generator for discrete data and a generator for continuous data, as well as a time series fully connected layer and a fully connected layer, the high-order features of the latent space output by the generator for discrete data and the generator for continuous data are input to the time series connected layer, the output of the time series fully connected layer is input to the fully connected layer, and the output of the fully connected layer is input to the comparator; The discriminator includes a continuous data discriminator and a discrete data discriminator. The output of the comparator is decoded and input into the discriminator to output a label for distinguishing synthetic data from real data.

2. The data privacy protection system according to claim 1, wherein: The coupled generative adversarial network includes a BLSTM network for continuous data and a BLSTM network for discrete data coupled to each other, and also includes a fully connected layer for continuous data and a fully connected layer for discrete data; wherein, The BLSTM network of continuous data is used as a generator of continuous data, and continuous random noise is input to obtain The hidden layer output at the moment is input to the fully connected layer of continuous data; The BLSTM network of discrete data is used as a generator of discrete data and is input with discrete random noise to obtain The hidden layer output at the moment is input to the fully connected layer of the discrete data; exist At time t, the inputs of the BLSTM network for continuous data and the BLSTM network for discrete data both include: The hidden layer output of the BLSTM network for continuous data at time t, and The hidden layer output of the BLSTM network of discrete data at the moment is obtained The hidden layer output at time t.

3. The data privacy protection system according to claim 2, wherein: The inputs of the continuous data encoder and the discrete data encoder are respectively The real data at the time, and the time The corresponding decoder outputs the reconstructed data and hidden state tensor when ; The outputs of the continuous data encoder and the discrete data encoder are sampled respectively to obtain the time The latent space high-order features are input into the corresponding decoders respectively; The decoders are based on time The hidden state tensor of The latent space high-order features, decoding generation moment The reconstruction data, The comparator is based on the time Real data and time The reconstructed data is used for comparative learning.

4. The data privacy protection system according to claim 3, wherein: The temporal fully connected layer outputs a tensor of continuous data and a tensor of discrete data, which are input into the fully connected layer; the fully connected layer outputs the final latent space high-order feature representation.

5. The data privacy protection system according to claim 4, wherein: The input vector of the time series fully connected layer is two tensors of the same dimension, one continuous and one discrete. After being spliced ​​and input to the fully connected layer, the output vector is split into a tensor of continuous data and a tensor of discrete data of equal length.

6. The data privacy protection system according to claim 1, wherein: The patient's vital signs data include at least one of heart rate, blood pressure, blood oxygen saturation, and respiratory rate; the medical intervention measures include binary status identifiers of ventilator use, antihypertensive drug use, and pressor drug use.

7. A data privacy protection method, implemented based on the data privacy protection system according to any one of claims 1 to 6, characterized in that The steps include: Input continuous random noise and historical vital signs data into the continuous data generator to obtain The hidden layer output at time ; Input discrete random noise and medical intervention status at historical moments into the discrete data generator to obtain The hidden layer output at time ; Will Input the continuous fully connected layer to get the continuous real data The corresponding high-order features of the latent space ;Will Input discrete fully connected layer to get discrete real data The corresponding high-order features of the latent space ; The high-order features of the latent space and Input to the time series fully connected layer, The output at each moment is a tensor group , split to get the tensor of continuous data and discrete data tensors ,in Indicates from Until the moment The time step of the moment; Will 、 Input to the fully connected layer to output the final latent space high-order feature representation and to the comparator; Use the encoder for continuous data and the encoder for discrete data to complete the original data respectively. The encoding is sampled by the continuous data sampling module and the discrete data sampling module to obtain the moment with the same feature dimension The latent space high-order features are decoded to obtain the reconstructed data, which is input to the comparator; After comparative learning by the comparator, they are encoded separately and input into the continuous data discriminator and the discrete data discriminator to output labels for distinguishing synthetic data from real data.

8. The data privacy protection method according to claim 7, wherein: By coupling a bidirectional long short-term memory network, the temporal correlation between vital sign characteristics and medical intervention status characteristics is jointly modeled to generate synthetic data that conforms to medical logic.

9. The data privacy protection method according to claim 7, wherein: The encoder of the continuous data captures the causal relationship between blood pressure changes and drug intervention through variational lower bound loss and contrastive learning loss; the temporal fully connected layer of the coupled generative adversarial network dynamically associates the vital sign data at the current moment with the medical intervention status at the previous moment.

10. The data privacy protection method according to claim 7, wherein: The encoder of the continuous data and the encoder of the discrete data are constructed into a dual variational autoencoder, and the loss function of the dual variational autoencoder is: variational lower bound loss , feature matching loss or contrastive learning loss .

Images