Methods and systems for generating privacy-preserving longitudinal synthetic data
The use of a GAN with a privacy loss function generates non-identifiable synthetic longitudinal data, addressing privacy concerns and enhancing data sharing in healthcare by producing clinically useful, high-quality data.
Patent Information
- Application Number
- PCT/EP2024/062106
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-10-30
- Filing Date
- 2024-05-02
- Publication Date
- 2024-11-07
AI Technical Summary
The healthcare sector faces challenges in accessing and sharing longitudinal patient data due to privacy concerns, with existing solutions for synthetic data generation not adequately addressing the need for privacy-preserving and clinically useful data.
A method and system using a trained generative adversarial network (GAN) to generate non-identifiable synthetic longitudinal data points, employing a privacy loss function to minimize the risk of data identification, thereby creating high-quality, privacy-preserving data suitable for healthcare analysis.
The solution effectively generates high-quality, privacy-preserving synthetic longitudinal data that is clinically useful, overcoming the limitations of existing data sharing restrictions and enhancing research reproducibility in healthcare.
Smart Images

Figure EP2024062106_07112024_PF_FP_ABST
Abstract
Description
METHODS AND SYSTEMS FOR GENERATING PRIVACY-PRESERVING LONGITUDINAL SYNTHETIC DATA Field of the Disclosure
[0001] The present disclosure relates generally to methods, systems and computer program products for generating and / or providing anonymous synthetic longitudinal data. Background
[0002] There is a growing need for massive longitudinal datasets, in many different fields, for a variety of purposes including both extrapolation of trends and other information, and for training machine learning models. The healthcare sector is just one of these many different fields in which massive amounts of longitudinal data are being used to make discoveries and improve outcomes. But there is always demand for more and more data.
[0003] The healthcare sector provides an additional challenge that must be overcome when analyzing longitudinal data, namely the issue of privacy. The challenges of too little data, and anonymized data, are significant hurdles. Indeed, one difficulty of advancing innovative ideas in the healthcare sector stems from the obstacle of scarce data availability. Before innovation can improve lives, it must be tested on large amounts of patient data. Accessing such data without violating patient privacy is often challenging. Regulations such as the General Data Protection Regulation (GDPR) and Health Insurance Portability and Accountability Act (HIPAA) safeguard patient privacy but have also led to arduous data-sharing processes.
[0004] Currently, the GDPR is the strictest privacy law globally, and it applies to all data under its definition of personal data. Under the GDPR, sharing personal data can only be done with individual informed consent or strict exceptions. To minimize the risk of violating these regulations, data sharing often relies on extensive contracts and individual analysis. These contracts and agreements tend to restrict the broader use of data outside the specific study, thereby limiting the reproducibility of studies using patient data.
[0005] Synthetic data (SD) generators have emerged as a potential solution to these challenges. These generators create text, image, tabular, and time-series data with similar statistical properties to the original data while having the potential to preserve patient privacy. As anonymous data isnot subject to the GDPR, it can be shared more quickly. However, regulations such as the GDPR typically do not provide a mathematical definition of anonymous data. Moreover, the central question is whether SD can be both privacy-preserving and clinically useful in healthcare. The industry is, therefore, actively exploring whether SD can fulfill its promises. Despite existing research investigating the generation of longitudinal data and privacy-preserving static data, the generation of privacy-preserving longitudinal data remains unexplored. Summary of the Disclosure
[0006] Accordingly, there is a continued need for systems and methods that generate privacy- preserving synthetic longitudinal data.
[0007] Various embodiments and implementations herein are directed to a method and system configured to generate privacy-preserving synthetic longitudinal data using a trained generative adversarial network (GAN). According to an embodiment, a GAN is trained to generate non- identifiable synthetic longitudinal data comprising a plurality of synthetic longitudinal data points using a training dataset comprising a plurality of longitudinal data points, the dataset comprising non-synthetic identifiable data. Each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points. During training the GAN utilizes a privacy loss function when generating the synthetic longitudinal data, the privacy loss function configured to compare a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point, and further configured to minimize the loss function such that the generated synthetic longitudinal data point is non-identifiable.
[0008] Generally, in one aspect, a method for training a generative adversarial network (GAN) to generate non-identifiable synthetic longitudinal data is provided. The method includes: (i) providing a training dataset comprising a plurality of longitudinal data points, wherein the dataset comprises non-synthetic identifiable data; and training the GAN, using the provided training dataset, to generate a synthetic longitudinal dataset comprising a plurality of synthetic longitudinal data points; wherein each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points; wherein the generative adversarial network utilizes a privacy loss function when generating the synthetic longitudinaldata, the privacy loss function configured to compare a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point, and further configured to minimize the loss function such that the generated synthetic longitudinal data point is non-identifiable.
[0009] According to an embodiment, the non-synthetic identifiable data comprises identifiable patient health data.
[0010] According to an embodiment, comparing a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point comprises calculating a distance between the generated synthetic longitudinal data and the respective longitudinal data point.
[0011] According to an embodiment, each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points using a predetermined noise vector.
[0012] According to another aspect is a system for training a generative adversarial network (GAN) to generate non-identifiable synthetic longitudinal data. The system includes: a dataset comprising a plurality of longitudinal data points, wherein the dataset comprises non-synthetic identifiable data; and one or more computer processors utilizing a generative adversarial network (GAN) configured to be trained to generate non-identifiable synthetic longitudinal data, wherein the GAN comprises a privacy loss function configured to, during training, compare a generated synthetic longitudinal data point to a respective identifiable non-synthetic longitudinal data point utilized to generate the synthetic longitudinal data point, and further configured to minimize the loss function such that the generated synthetic longitudinal data point is not identifiable.
[0013] According to an embodiment, the non-synthetic identifiable data comprises identifiable patient health data.
[0014] According to an embodiment, comparing a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point comprises calculating a distance between the generated synthetic longitudinal data and the respective longitudinal data point.
[0015] According to an embodiment, each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points using a predetermined noise vector.
[0016] According to another aspect is a system for providing anonymous synthetic longitudinal data to a user. The system includes: one or more computer processors comprising: a user interface configured to receive a query from a user or user computing device requesting a synthetic longitudinal dataset; a trained generative adversarial network (GAN) configured to generate the requested synthetic longitudinal dataset, wherein the trained GAN comprises a privacy loss function configured to compare a generated synthetic longitudinal data point to a respective identifiable non-synthetic longitudinal data point utilized to generate the synthetic longitudinal data point, and further configured to minimize the loss function such that the generated synthetic longitudinal data point is not identifiable; and a delivery component configured to provide the generated synthetic longitudinal dataset to the user or user computing device. For example, the trained GAN may be trained according to the system or method for training a GAN described above.
[0017] According to an embodiment, comparing a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point comprises calculating a distance between the generated synthetic longitudinal data and the respective longitudinal data point.
[0018] According to an embodiment, each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points using a predetermined noise vector.
[0019] According to another aspect is a method for providing anonymous synthetic longitudinal data to a user. This method comprises the step of receiving a query from a user or user computing device requesting a synthetic longitudinal dataset. The receiving is by means of a user interface. The method comprises the step of generating the requested synthetic longitudinal dataset by means of a trained generative adversarial network (GAN). The trained GAN comprises a privacy loss function. For example, the trained GAN may be trained according to the system or method for training a GAN described above. The method further comprises the step of comparing, by means of the trained GAN, a generated synthetic longitudinal data point to a respective identifiable non-synthetic longitudinal data point utilized to generate the synthetic longitudinal data point. The method comprises the step of minimizing, by means of the trained GAN, the loss function such that the generated synthetic longitudinal data point is not identifiable and the step of providing, by means of a delivery component, the generated synthetic longitudinal dataset to the user or user computing device.
[0020] It should be appreciated that all combinations of the foregoing concepts and additional concepts discussed in greater detail below (provided such concepts are not mutually inconsistent) are contemplated as being part of the inventive subject matter disclosed herein. In particular, all combinations of claimed subject matter appearing at the end of this disclosure are contemplated as being part of the inventive subject matter disclosed herein. It should also be appreciated that terminology explicitly employed herein that also may appear in any disclosure incorporated by reference should be accorded a meaning most consistent with the particular concepts disclosed herein.
[0021] These and other aspects of the various embodiments will be apparent from and elucidated with reference to the embodiment(s) described hereinafter. Brief Description of the Drawings
[0022] In the drawings, like reference characters generally refer to the same parts throughout the different views. The figures showing features and ways of implementing various embodiments and are not to be construed as being limiting to other possible embodiments falling within the scope of the attached claims. Also, the drawings are not necessarily to scale, emphasis instead generally being placed upon illustrating the principles of the various embodiments.
[0023] FIG. 1 is a flowchart of a method for training a generative adversarial network (GAN) to generate non-identifiable synthetic longitudinal data, in accordance with an embodiment.
[0024] FIG. 2 is a schematic representation of a synthetic longitudinal data system, in accordance with an embodiment.
[0025] FIG. 3 is a schematic representation of an architecture of a generative adversarial network, in accordance with an embodiment.Detailed Description of Embodiments
[0026] The present disclosure describes various embodiments of a system and method configured to generate privacy-preserving synthetic longitudinal data. More generally, Applicant has recognized and appreciated that it would be beneficial to generate synthetic longitudinal data that is both privacy-preserving and clinically useful in healthcare. Accordingly, a synthetic longitudinal data system generates privacy-preserving synthetic longitudinal data using a trained generative adversarial network (GAN). According to an embodiment, a GAN is trained to generate non-identifiable synthetic longitudinal data comprising a plurality of synthetic longitudinal data points using a training dataset comprising a plurality of longitudinal data points, the dataset comprising non-synthetic identifiable data. Each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points. During training the GAN utilizes a privacy loss function when generating the synthetic longitudinal data, the privacy loss function configured to compare a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point, and further configured to minimize the loss function such that the generated synthetic longitudinal data point is non-identifiable.
[0027] Generating privacy-preserving longitudinal SD is a complex problem, so narrowing down exactly what the SD generator should be capable of is important. Longitudinal healthcare data consists of static data and longitudinal data. Given a dataset D, let S be the static vector space, and let X be the temporal vector space. In addition, let S ∈ S and X ∈ X be random vectors that can be instantiated with values s and x, respectively. In this setting, data takes the form of a tuple (S, X1:T), where T denotes the sequence length of the temporal vector X. Following this notation, a dataset of size N is denoted
[0028] An SD generator needs to learn a distribution ^̂(S, X1:T) over a training dataset D approximating the actual distribution p(S, X1:T), such that if learned successfully, the generator can generate data according to the distribution of the training data. However, it is possible to simplify the problem by using the autoregressive properties of the time-series data. Specifically, the joint data distribution p(S, X1:T) can be split up over the time domain through autoregressive decomposition, formalized as p(S, X1:T) = p(S) ^tp(Xt|S, X1:t−1). This results in a simpler objectiveto learn than the aforementioned distribution, as the generator needs to learn ^̂(Xt|S, X1:t−1) approximating p(Xt|S, X1:t−1) for any t.
[0029] Apart from formalizing the SD quality objectives of a generator, the objectives concerning privacy in SD should also be formalized. While the GDPR defines anonymous data as “information which does not relate to an identified or identifiable natural person,” it is not always clear whether data is anonymous. Information disclosed in the future can potentially be utilized to identify people in datasets that were previously released as anonymous. It is, therefore, unlikely that governments will specify thresholds for acceptable privacy, leaving it up to the industry to make risk-managed judgment calls on what is acceptable under current law. Hence, the development of models that can be fine-tuned for specific privacy applications will have a significant impact on healthcare research. In addition, the trade-offs between quality and privacy should be clear so that parties can estimate whether SD can be applicable to their problem.
[0030] The methods and systems disclosed or otherwise envisioned herein provide a system for generating high-quality SD that preserves privacy. The system comprises a new neural architecture, called Time-ADS-GAN, created to generate privacy-preserving longitudinal SD, blending and then modifying the TimeGAN and ADS-GAN architectures in a novel and non- obvious way. Rigorous evaluation of Time-ADS-GAN demonstrates that the novel methods and systems are able to generate high-quality SD that preserves privacy and is useful for healthcare analysis.
[0031] According to an embodiment, the systems and methods described or otherwise envisioned herein can, in some non-limiting embodiments, be implemented as an element for a commercial product to generate privacy-preserving synthetic longitudinal data. According to an embodiment, the systems and methods described or otherwise envisioned herein can be implemented as a service configured to provide one or more datasets comprising privacy- preserving synthetic longitudinal data. According to an embodiment, the systems and methods described or otherwise envisioned herein can be implemented as a generative adversarial network (GAN) which is trained to generate privacy-preserving synthetic longitudinal data. However, the disclosure is not limited to these devices or systems, and thus the disclosure and embodiments disclosed herein can encompass any device or system that generates or utilizes privacy-preserving synthetic longitudinal data.
[0032] Referring to FIG. 1, in one embodiment, is a flowchart of a method 100 for generating privacy-preserving synthetic longitudinal data using a longitudinal data system 200 comprising a generative adversarial network (GAN). The methods described in connection with the figures are provided as examples only, and shall be understood not limit the scope of the disclosure. The longitudinal data system can be any of the systems described or otherwise envisioned herein. The longitudinal data system can be a single system or multiple different systems.
[0033] At step 110 of the method, a longitudinal data system 200 is provided. Referring to an embodiment of a longitudinal data system 200 as depicted in FIG. 2, for example, the system comprises one or more of a processor 220, memory 230, user interface 240, communications interface 250, and storage 260, interconnected via one or more system buses 212. It will be understood that FIG.2 constitutes, in some respects, an abstraction and that the actual organization of the components of the system 200 may be different and more complex than illustrated. Additionally, longitudinal data system 200 can be any of the systems described or otherwise envisioned herein. Other elements and components of longitudinal data system 200 are disclosed and / or envisioned elsewhere herein.
[0034] At step 120 of the method, the longitudinal data system receives a request for privacy- preserving synthetic longitudinal data. The request can be received via a variety of different mechanisms. For example, the request can be received from a user via a user interface of the longitudinal data system or another system in communication with the longitudinal data system, or the request can be received from a user computing device that is a component of the longitudinal data system or another system in communication with the longitudinal data system. The request can include one or more parameters for the synthetic longitudinal dataset. The request may also provide other specifications for the synthetic longitudinal dataset. The user may have an intended use for the synthetic longitudinal dataset, which may or may not affect the generation of that dataset.
[0035] At step 130 of the method, the longitudinal data system generates the privacy-preserving synthetic longitudinal data in response to the received request. According to an embodiment, the longitudinal data system utilizes a trained neural architecture called Time-ADS-GAN, created to generate the privacy-preserving longitudinal synthetic data. As described or otherwise envisionedherein, the Time-ADS-GAN blends and modifies the TimeGAN and ADS-GAN architectures in a novel and non-obvious way to be able to generate the privacy-preserving longitudinal synthetic data.
[0036] Once generated, the privacy-preserving synthetic longitudinal data may be utilized immediately, or may be stored in local or remote storage for use in further steps of the method.
[0037] At step 140 of the method, the generated privacy-preserving synthetic longitudinal data is provided to the user via the user interface of the longitudinal data system or a user interface of another system in communication with the longitudinal data system. The generated data may be provided to the requesting user computing device. The generated data may be provided in any format. For example, the generated data may be provided via a physical storage device or may be provided via access to a local or remote database storing the generated data, among other mechanisms. The generated data may be provided with other information about the dataset, such as size or other metrics or parameters of the dataset.
[0038] To generate privacy-preserving synthetic longitudinal data that is still useful for healthcare applications, the generative adversarial network must be trained to do so. Accordingly, at step 150 of the method, the generative adversarial network is trained. According to an embodiment, training 150 comprises a series of steps.
[0039] According to an embodiment, the generative adversarial network is trained at any point prior to being used as a trained generative adversarial network utilized to generate privacy- preserving synthetic longitudinal data. For example, the generative adversarial network may be trained immediately before generating the requested data. As another example, the generative adversarial network may be trained in advance as a component of a commercial product, such as a local or remote service accessed via a computer network, that is utilized by one or more users to generate privacy-preserving synthetic longitudinal data. According to an embodiment, the generative adversarial network is initially trained locally or remotely, and is then re-trained or fine- tuned one or more times with new or additional training data at one or more future dates. Many other options are possible.
[0040] At step 150 of the method, a training dataset is provided, received, or otherwise obtained. The training dataset is composed at least partially of non-synthetic data and comprises aplurality of longitudinal data points and optionally static data points. According to an embodiment, particularly in the healthcare setting, the training data comprises identifiable data. For example, the training dataset may comprise patient health records.
[0041] The training dataset may be provided, received, or otherwise obtained from any source that comprises the requisite data. The training dataset may be received or obtained from one or a plurality of different sources. According to an embodiment, the training dataset is received from, retrieved from, or otherwise obtained in whole or in part from an electronic medical record database or system 270. The EMR database or system may be local or remote. The EMR database or system may be a component of the longitudinal data system 200 or may be in local and / or remote communication with the longitudinal data system. The received training dataset may be utilized immediately, or may be stored in local or remote storage for use in further steps of the method.
[0042] According to an embodiment, the longitudinal data system may comprise a data pre- processor or similar component or algorithm configured to process the received training data. For example, the data pre-processor analyzes the training data to remove noise, bias, errors, and other potential issues. The data pre-processor may also analyze the input data to remove low-quality data. Many other forms of data pre-processing or data point identification and / or extraction are possible.
[0043] At step 154 of the method, the system trains the generative adversarial network, which will be the trained GAN utilized to generate the privacy-preserving synthetic longitudinal data. The network is trained using the training dataset according to known methods for training a network. However, as described or otherwise envisioned herein, the architecture of the generative adversarial network is novel and thus the training of the network is novel.
[0044] Referring to FIG.3, in one embodiment, is the architecture of the generative adversarial network 300. The generative adversarial network comprises a plurality of components that are trained to generate the privacy-preserving synthetic longitudinal data, as described or otherwise envisioned herein.
[0045] The generative adversarial network 300 comprises a plurality of components, including an embedding component 310, a recovery component 320, a generator component 330, a discriminator component 340, and a supervisor component 350. According to an embodiment,these components are separate components. According to another embodiment, two or more of the components may be combined, programmed, or designed into a single component that may perform each of the respective components.
[0046] According to an embodiment, the embedding component 310 and recovery component 320 form an autoencoder which provides the generative adversarial network with an easy to learn latent space and allows the model to generate the data in a potentially lower dimensional space. Let ^s, ^xrepresent the represent the latent spaces corresponding to the features spaces S and X. Given a dataset D, let S be the static vector space, and let X be the temporal vector space. In addition, let S ∈ S and X ∈ X be random vectors that can be instantiated with values s and x, respectively. In this setting, data takes the form of a tuple (S, X1:T), where T denotes the sequence length of the temporal vector X. Following this notation, a dataset of size N is denoted as D = {(sn,
[0047] According to an embodiment, the embedding function e takes in the feature space and maps it to the embedding space, formally denoted as S x ΠtX → ^s× Πt^x, therewith producing latent codes hS, h1:T = e(s,X1:T). The embedded static features are used to embed the temporal features. The embedding function e is implemented through a recurrent neural net:where eS : S → ^s and ex : ^x, × X × Cex → ^x represent the embedding functions for the static and temporal features respectively, and ceX ,t−1represents the carry state (often called the hidden state) of the autoregressive cell. Although it is sometimes assumed that the output of eX is equal to the carry state, this does not need to be, i.e., if the autoregressive cell used is an LSTM. CeX represents the carry state feature space, subscripted to denote the function it is referring to.
[0048] Recovery component 320 functions very similarly but in reverse. Formally, the recovery function r : ^s× Πt^x→ ^ × ΠtX takes the embedded static and temporal features and maps them back to their feature spaces ^̃, ^^1:T = r(ℎ^S, ℎ^1:T). The ~ symbol notation above a variable denotes that the variable can be either constructed using a real sample or generated using a random prior. In this case, the recovery module can have as input an embedding of a real sample or agenerated embedding of the generator. Here, r is implemented through a feedforward network at each step:where rX: ^S→ S and rX: ^x→ X are recovery networks for the static and temporal features respectively. However, the recovery function rX can better be described as:(Eq.3) where rX : ^x × Ce → X is the formalization of rX and c^rX,t-1 is again representing the carry state.
[0049] Generator component 330 synthesizes data in the embedding space, on which the discriminator also discriminates the samples.
[0050] Following is described the generator component in previous architectures (as will be shown, the generator component is modified for the present invention). For previous architectures, let ƵS, ƵXdenote the prior spaces for the generator. The generating function g : ΠtƵx→ ^s× Πt^x takes a vector from both aforementioned priors and synthesizes a sample in the embedding space, denoted h^S, h^1:T = g(ZS,Z1:T). Theˆ above a variable denotes that the variable is synthetically generated. Again, g is implemented through a recurrent network,where gS : ƵX → ^s and gx : ^S, × Ƶx × Cgx → ^x are generator networks for the static and temporal features respectively.
[0051] The changed generator, generator component 330, is also conditioned on the real data. The conditional data is concatenated with the random vectors before they are inserted in the generator networks. Hence the generator function g : (ƵS × Πt Ƶx) × (S × Πt X) → ^s × Πt ^x takes a tuple of tuples of static and temporal random and real vectors and generates synthetic latent codes h^S, h^1:T = gX(zS, z1:T, s, x1:T). g is implemented through a recurrent network:where gS : ƵS × S → ^s is a static feature generator network and gx : ^S × Ƶx × X × Cgx → ^x is a recurrent generator for temporal features.
[0052] Like the recovery module, discriminator component 340 operates from the embedding space. The discriminator function d : ^s × Πt ^x → [0,1] × Πt [0,1] transforms the embedding space to a classification "^S, y^1:T= d(ℎ^S, h^1:T). And d is implemented through a bidirectional recurrent network with a feedforward output layer: (Eq.6) where $%t= '%x(ℎ^S, h^t, (̃)% *,+,^) and $%t= ' %x (ℎ^S, h^t, (̃)%*,+,^) denote bidirectional outputs of the recurrent neural nets, '%, '%are recurrent functions, and dS, dX are classification functions.
[0053] Supervisor component 350 takes in the embedding space and uses the input to predict the next time step in the embedding space. The supervisor function s : ^s× Πt^x→ Πt^xtransforms the embedding space of both the static and temporal features into an embedding space of only the temporal features (the next time step). Logically, the static features do not need to be transformed, as they are the same at every time step. Like most other modules, s is implemented through a recurrent network with a feedforward output layer:(Eq.7) where ℎ^t,sup represents the supervised (predicted) embedded sample or random prior at time step t, and (̃s,t-1 represents the carry state of the recurrent cell from time-step t − 1. ℎ^t,sup should not be confused with ℎ^which has not been processed by the supervisor. Notably, there are no architectural restrictions on any of the modules as long as they are autoregressive.
[0054] According to an embodiment, the generative adversarial network 300 comprises a plurality of loss functions. The components are connected through the different losses with which the network is trained. Four of these possible plurality of loss functions are discussed below.
[0055] Referring again to FIG. 3, there is the first loss function, reconstruction loss (.R) 362, which is used to introduce a reversible mapping from the feature space to the latent space and back to the feature space, much like an autoencoder. The loss is a mean squared error loss over the inputand output of the embedding and recovery components. Upon backpropagation, the reconstruction loss influences the weights of both the embedding and recovery modules.
[0056] The second loss function, unsupervised loss (.U) 366, is the basis of the traditional GAN framework. The generator generates features in the latent space, and the embedding module embeds real samples into the latent space. The discriminator then discriminates these samples with a binary cross-entropy loss and backpropagates this towards the generator and the discriminator. In contrast to the standard GAN framework, the discriminator gets three batches of data, a batch of real embedded data, a batch of fake generated data, and a batch of fake generated and supervised data. This last component helps to train the generator to copy the embedding learned by the embedding module. Note that the discriminator loss is not backpropagated towards the embedding module.
[0057] A loss is introduced to aid the generator in learning to generate in the latent space. This third loss function, supervised loss (.S) 368, trains the supervisor in a closed-loop fashion on real samples. The supervised loss is implemented using the mean squared error loss between the prediction and the actual next time step of the embedding. The supervised loss is only backpropagated towards the supervisor module.
[0058] The fourth loss function is identifiability loss function (.I) 364, which implemented to train a model to optimize privacy and generate a privacy-protecting synthetic dataset. For static features, the loss remains the same and it can be weighted with a weight vector wSspecific for the static features. For the temporal features, it is logical to span the loss over time and weigh the features with a weight vector wXspecific for temporal features. Similarly to the static features, the distance between samples can be calculated using the Euclidean distance. A different distance function can also be used if that would be more suitable for a specific use case or data source. To keep the formalization modular, a distance function δ is introduced for calculating the distance between an original sample x1:Tand an SD sample x31:T. In the rest of the research, the Euclidean distance will be used in implementations. Based on these decisions, the identifiability loss function (.I) 364 can be defined as:
[0059] The identifiability loss is calculated on the embedded real samples and the supervised generated samples. Hence, during the training of the generator, the identifiability loss is backpropagated through the generator and supervisor modules. The identifiability loss does not influence the embedding or the recovery modules.
[0060] Returning to method 100 in FIG.1, at step 156 of the method, the generative adversarial network is stored for future use. According to an embodiment, the model may be stored in local or remote storage.
[0061] Referring to FIG. 2 is a schematic representation of a longitudinal data system 200. System 200 may be any of the systems described or otherwise envisioned herein, and may comprise any of the components described or otherwise envisioned herein. It will be understood that FIG. 2 constitutes, in some respects, an abstraction and that the actual organization of the components of the system 200 may be different and more complex than illustrated.
[0062] According to an embodiment, system 200 comprises a processor 220 capable of executing instructions stored in memory 230 or storage 260 or otherwise processing data to, for example, perform one or more steps of the method. Processor 220 may be formed of one or multiple modules. Processor 220 may take any suitable form, including but not limited to a microprocessor, microcontroller, multiple microcontrollers, circuitry, field programmable gate array (FPGA), application-specific integrated circuit (ASIC), a single processor, or plural processors.
[0063] Memory 230 can take any suitable form, including a non-volatile memory and / or RAM. The memory 230 may include various memories such as, for example L1, L2, or L3 cache or system memory. As such, the memory 230 may include static random access memory (SRAM), dynamic RAM (DRAM), flash memory, read only memory (ROM), or other similar memory devices. The memory can store, among other things, an operating system. The RAM is used by the processor for the temporary storage of data. According to an embodiment, an operating system may contain code which, when executed by the processor, controls operation of one or more components of system 200. It will be apparent that, in embodiments where the processor implements one or more of the functions described herein in hardware, the software described as corresponding to such functionality in other embodiments may be omitted.
[0064] User interface 240 may include one or more devices for enabling communication with a user. The user interface can be any device or system that allows information to be conveyed and / or received, and may include a display, a mouse, and / or a keyboard for receiving user commands. In some embodiments, user interface 240 may include a command line interface or graphical user interface that may be presented to a remote terminal via communication interface 250. The user interface may be located with one or more other components of the system, or may located remote from the system and in communication via a wired and / or wireless communications network.
[0065] Communication interface 250 may include one or more devices for enabling communication with other hardware devices. For example, communication interface 250 may include a network interface card (NIC) configured to communicate according to the Ethernet protocol. Additionally, communication interface 250 may implement a TCP / IP stack for communication according to the TCP / IP protocols. Various alternative or additional hardware or configurations for communication interface 250 will be apparent.
[0066] Storage 260 may include one or more machine-readable storage media such as read- only memory (ROM), random-access memory (RAM), magnetic disk storage media, optical storage media, flash-memory devices, or similar storage media. In various embodiments, storage 260 may store instructions for execution by processor 220 or data upon which processor 220 may operate. For example, storage 260 may store an operating system 261 for controlling various operations of system 200.
[0067] It will be apparent that various information described as stored in storage 260 may be additionally or alternatively stored in memory 230. In this respect, memory 230 may also be considered to constitute a storage device and storage 260 may be considered a memory. Various other arrangements will be apparent. Further, memory 230 and storage 260 may both be considered to be non-transitory machine-readable media. As used herein, the term non-transitory will be understood to exclude transitory signals but to include all forms of storage, including both volatile and non-volatile memories.
[0068] While system 200 is shown as including one of each described component, the various components may be duplicated in various embodiments. For example, processor 220 may includemultiple microprocessors that are configured to independently execute the methods described herein or are configured to perform steps or subroutines of the methods described herein such that the multiple processors cooperate to achieve the functionality described herein. Further, where one or more components of system 200 is implemented in a cloud computing system, the various hardware components may belong to separate physical systems. For example, processor 220 may include a first processor in a first server and a second processor in a second server. Many other variations and configurations are possible.
[0069] According to an embodiment, the longitudinal data system comprises an electronic medical record (EMR) database or system 270. Alternatively, the EMR database or system 270 may be a local or remote database or system and thus the longitudinal data system may be in direct and / or indirect communication with the EMR database or system 270.
[0070] According to an embodiment, storage 260 of system 200 may store one or more algorithms, modules, and / or instructions to carry out one or more functions or steps of the methods described or otherwise envisioned herein. For example, the system may comprise, among other instructions or data, a training dataset 262, a trained generative adversarial network (GAN). 263, and / or reporting instructions 264.
[0071] According to an embodiment, a training dataset 262 is utilized to train a generative adversarial network. The training dataset is composed at least partially of non-synthetic data and comprises a plurality of longitudinal data points and optionally static data points. According to an embodiment, particularly in the healthcare setting, the training data comprises identifiable data. For example, the training dataset may comprise patient health records. The training dataset may be provided, received, or otherwise obtained from any source that comprises the requisite data. The training dataset may be received or obtained from one or a plurality of different sources. According to an embodiment, the training dataset is received from, retrieved from, or otherwise obtained in whole or in part from an electronic medical record database or system 270. The EMR database or system may be local or remote. The received training dataset may be utilized immediately, or may be stored in local or remote storage for use in further steps of the method.
[0072] According to an embodiment, the trained generative adversarial network 263 is trained to generate privacy-preserving synthetic longitudinal data. The generative adversarial network istrained using the training dataset according to known methods for training a network. However, as described or otherwise envisioned herein, the architecture of the generative adversarial network is novel and thus the training of the network is novel.
[0073] According to an embodiment, the reporting instructions 264 direct the system to generate and provide to a user via a user interface or computer network the generated privacy- preserving synthetic longitudinal data. The generated data may be provided to the user via any mechanism for display, visualization, or otherwise providing information via a user interface. According to an embodiment, the information may be communicated by wired and / or wireless communication to a user interface and / or to another device. For example, the system may communicate the information to a mobile phone, computer, laptop, wearable device, and / or any other device configured to allow display and / or other communication of the report. The user interface can be any device or system that allows information to be conveyed and / or received, and may include a display, a mouse, and / or a keyboard for receiving user commands.
[0074] Accordingly, within the context of the disclosure herein, aspects of the embodiments may take the form of a computer program product embodied in one or more non-transitory computer-readable media having computer readable program code embodied thereon. Thus, according to one embodiment is a non-transitory computer-readable storage medium comprising computer program code instructions which, when executed by a processor, enables the processor to carry out a method including: (i) providing a training dataset comprising a plurality of longitudinal data points, wherein the dataset comprises non-synthetic identifiable data; (ii) training the GAN, using the provided training dataset, to generate a synthetic longitudinal dataset comprising a plurality of synthetic longitudinal data points, where each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points, and where the generative adversarial network utilizes a privacy loss function when generating the synthetic longitudinal data, the privacy loss function configured to compare a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point, and further configured to minimize the loss function such that the generated synthetic longitudinal data point is non-identifiable.
[0075] Further, according to another embodiment is a non-transitory computer-readable storage medium comprising computer program code instructions which, when executed by a processor,enables the processor to carry out a method including: (i) receiving a query from a user or user computing device requesting a synthetic longitudinal dataset; (ii) generating the requested synthetic longitudinal dataset using a trained generative adversarial network (GAN), where the trained generative adversarial network (GAN) comprises a privacy loss function configured to, during training, compare a generated synthetic longitudinal data point to a respective identifiable non-synthetic longitudinal data point utilized to generate the synthetic longitudinal data point, and further configured to minimize the loss function such that the generated synthetic longitudinal data point is not identifiable; and (iii) providing the generated synthetic longitudinal dataset to the user or user computing device.
[0076] According to an embodiment, the longitudinal data system 200 is configured to generate synthetic data while being trained using actual data. This creates a privacy-preserving wall between the actual data input used to train the generative adversarial network and the synthetic data generated by the system. Accordingly, the training of the system with actual data and the generation of the synthetic data are processes that cannot be performed in a human mind, as the mind cannot preserve privacy. Indeed, shielding minds from the private information within the input data, by generating synthetic data, is a primary goal and thus outcome of the systems and methods described or otherwise envisioned herein.
[0077] According to an embodiment, the longitudinal data system 200 is configured to process millions of datapoints in the input data used to train the system, as well as to generate a privacy- preserving synthetic longitudinal dataset once the system is trained. For example, generating a functional and skilled trained system using an automated process such as training requires processing of millions of datapoints from input data and the generated features. This can require millions or billions of calculations to generate a novel trained system from those millions of datapoints and millions or billions of calculations. As a result, the trained system is novel and distinct based on the input data and parameters of the network, and thus improves the functioning of a data generation system, particularly one requiring privacy of the generated data. Generating a functional and skilled trained system comprises a process with a volume of calculation and analysis that a human brain cannot accomplish in a lifetime, or multiple lifetimes. By providing improved privacy preserving data generation system, this novel longitudinal data system has an enormous positive effect on data analytics such as healthcare, compared to prior art systems.
[0078] Critical to the applicability of synthetic data in the healthcare industry, for example, is whether research can be done with it. The reproduction of research using an in-patient deterioration dataset shows that Time-ADS-GAN is far superior in its performance compared to prior art systems. The prediction model trained on the synthetic data, created by Time-ADS-GAN, yielded results closely resembling those of the original prediction model. These results were unobtainable using prior art systems.
[0079] Thus, Time-ADS-GAN outperforms its predecessors and is able to generate more private, higher-quality synthetic data. The data generated using Time-ADS-GAN shows a significant improvement over prior art systems in reproduction studies, indicating that there is much potential for synthetic data to aid in speeding up the research process in the healthcare sector and potentially in many other sectors as well.
[0080] It should be appreciated that all combinations of the foregoing concepts and additional concepts discussed in greater detail below (provided such concepts are not mutually inconsistent) are contemplated as being part of the inventive subject matter disclosed herein. In particular, all combinations of claimed subject matter appearing at the end of this disclosure are contemplated as being part of the inventive subject matter disclosed herein. It should also be appreciated that terminology explicitly employed herein that also may appear in any disclosure incorporated by reference should be accorded a meaning most consistent with the particular concepts disclosed herein.
[0081] All definitions, as defined and used herein, should be understood to control over dictionary definitions, definitions in documents incorporated by reference, and / or ordinary meanings of the defined terms.
[0082] The indefinite articles “a” and “an,” as used herein in the specification and in the claims, unless clearly indicated to the contrary, should be understood to mean “at least one.”
[0083] The phrase “and / or,” as used herein in the specification and in the claims, should be understood to mean “either or both” of the elements so conjoined, i.e., elements that are conjunctively present in some cases and disjunctively present in other cases. Multiple elements listed with “and / or” should be construed in the same fashion, i.e., “one or more” of the elements so conjoined. Other elements may optionally be present other than the elements specificallyidentified by the “and / or” clause, whether related or unrelated to those elements specifically identified.
[0084] As used herein in the specification and in the claims, the phrase “at least one,” in reference to a list of one or more elements, should be understood to mean at least one element selected from any one or more of the elements in the list of elements, but not necessarily including at least one of each and every element specifically listed within the list of elements and not excluding any combinations of elements in the list of elements. This definition also allows that elements may optionally be present other than the elements specifically identified within the list of elements to which the phrase “at least one” refers, whether related or unrelated to those elements specifically identified.
[0085] As used herein, although the terms first, second, third, etc. may be used herein to describe various elements or components, these elements or components should not be limited by these terms. These terms are only used to distinguish one element or component from another element or component. Thus, a first element or component discussed below could be termed a second element or component without departing from the teachings of the inventive concept.
[0086] Unless otherwise noted, when an element or component is said to be “connected to,” “coupled to,” or “adjacent to” another element or component, it will be understood that the element or component can be directly connected or coupled to the other element or component, or intervening elements or components may be present. That is, these and similar terms encompass cases where one or more intermediate elements or components may be employed to connect two elements or components. However, when an element or component is said to be “directly connected” to another element or component, this encompasses only cases where the two elements or components are connected to each other without any intermediate or intervening elements or components.
[0087] In the claims, as well as in the specification above, all transitional phrases such as “comprising,” “including,” “carrying,” “having,” “containing,” “involving,” “holding,” “composed of,” and the like are to be understood to be open-ended, i.e., to mean including but not limited to. Only the transitional phrases “consisting of” and “consisting essentially of” shall be closed or semi-closed transitional phrases, respectively.
[0088] It should also be understood that, unless clearly indicated to the contrary, in any methods claimed herein that include more than one step or act, the order of the steps or acts of the method is not necessarily limited to the order in which the steps or acts of the method are recited.
[0089] The above-described examples of the described subject matter can be implemented in any of numerous ways. For example, some aspects can be implemented using hardware, software or a combination thereof. When any aspect is implemented at least in part in software, the software code can be executed on any suitable processor or collection of processors, whether provided in a single device or computer or distributed among multiple devices / computers.
[0090] The present disclosure can be implemented as a system, a method, and / or a computer program product at any possible technical detail level of integration. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present disclosure.
[0091] The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium can be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium comprises the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.
[0092] Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local areanetwork, a wide area network and / or a wireless network. The network can comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.
[0093] Computer readable program instructions for carrying out operations of the present disclosure can be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, comprising an object oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the “C” programming language or similar programming languages. The computer readable program instructions can execute entirely on the user’s computer, partly on the user’s computer, as a stand- alone software package, partly on the user’s computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, comprising a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider). In some examples, electronic circuitry comprising, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) can execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present disclosure.
[0094] Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to examples of the disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer readable program instructions.
[0100] The computer readable program instructions can be provided to a processor of a, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions can also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture comprising instructions which implement aspects of the function / act specified in the flowchart and / or block diagram or blocks.
[0101] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.
[0102] The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various examples of the present disclosure. In this regard, each block in the flowchart or block diagrams can represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks can occur out of the order noted in the Figures. For example, two blocks shown in succession can, in fact, be executed substantially concurrently, or the blocks can sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.
[0103] Other implementations are within the scope of the following claims and other claims to which the applicant can be entitled.
[0104] While several inventive embodiments have been described and illustrated herein, those of ordinary skill in the art will readily envision a variety of other means and / or structures for performing the function and / or obtaining the results and / or one or more of the advantages described herein, and each of such variations and / or modifications is deemed to be within the scope of the inventive embodiments described herein. More generally, those skilled in the art will readily appreciate that all parameters, dimensions, materials, and configurations described herein are meant to be exemplary and that the actual parameters, dimensions, materials, and / or configurations will depend upon the specific application or applications for which the inventive teachings is / are used. Those skilled in the art will recognize, or be able to ascertain using no more than routine experimentation, many equivalents to the specific inventive embodiments described herein. It is, therefore, to be understood that the foregoing embodiments are presented by way of example only and that, within the scope of the appended claims and equivalents thereto, inventive embodiments may be practiced otherwise than as specifically described and claimed. Inventive embodiments of the present disclosure are directed to each individual feature, system, article, material, kit, and / or method described herein. In addition, any combination of two or more such features, systems, articles, materials, kits, and / or methods, if such features, systems, articles, materials, kits, and / or methods are not mutually inconsistent, is included within the inventive scope of the present disclosure.
Claims
Claims What is claimed is:
1. A method (100) for training a generative adversarial network (GAN) to generate non-identifiable synthetic longitudinal data, comprising: providing (152) a training dataset comprising a plurality of longitudinal data points, wherein the dataset comprises non-synthetic identifiable data; and training (154) the GAN, using the provided training dataset, to generate a synthetic longitudinal dataset comprising a plurality of synthetic longitudinal data points; wherein each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points; wherein the generative adversarial network utilizes a privacy loss function when generating the synthetic longitudinal data, the privacy loss function configured to compare a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point, and further configured to minimize the loss function such that the generated synthetic longitudinal data point is non-identifiable.
2. The method of claim 1, wherein the non-synthetic identifiable data comprises identifiable patient health data.
3. The method of any of claims 1-2, wherein comparing a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point comprises calculating a distance between the generated synthetic longitudinal data and the respective longitudinal data point.
4. The method of any of claims 1-3, wherein each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points using a predetermined noise vector.
5. The method of any of claims 1-4, wherein the generative adversarial network utilizes a reconstruction loss function, an unsupervised loss function, and a supervised loss function when generating the synthetic longitudinal data.
6. A system (200) for training a generative adversarial network (GAN) to generate non-identifiable synthetic longitudinal data, comprising: a dataset (262) comprising a plurality of longitudinal data points, wherein the dataset comprises non-synthetic identifiable data; and one or more computer processors (220) utilizing a generative adversarial network (GAN) (263) configured to be trained to generate non-identifiable synthetic longitudinal data, wherein the generative adversarial network (GAN) comprises a privacy loss function configured to, during training, compare a generated synthetic longitudinal data point to a respective identifiable non-synthetic longitudinal data point utilized to generate the synthetic longitudinal data point, and further configured to minimize the loss function such that the generated synthetic longitudinal data point is not identifiable.
7. The system of claim 6, wherein the non-synthetic identifiable data comprises identifiable patient health data.
8. The system of any of claim 6-7, wherein comparing a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point comprises calculating a distance between the generated synthetic longitudinal data and the respective longitudinal data point.
9. The system of any of claims 6-8, wherein each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points using a predetermined noise vector.
10. The system of any of claims 6-9, wherein the generative adversarial network utilizes a reconstruction loss function, an unsupervised loss function, and a supervised loss function when generating the synthetic longitudinal data.
11. A system (220) for providing anonymous synthetic longitudinal data to a user, comprising: one or more computer processors (220); and a memory (230) storing instructions for: a user interface (240) configured to receive a query from a user or user computing device requesting a synthetic longitudinal dataset; a trained generative adversarial network (GAN) (263), in particular trained according to the method according to any of claims 1-5, configured to generate the requested synthetic longitudinal dataset, wherein the trained GAN comprises a privacy loss function configured to compare a generated synthetic longitudinal data point to a respective identifiable non-synthetic longitudinal data point utilized to generate the synthetic longitudinal data point, and further configured to minimize the loss function such that the generated synthetic longitudinal data point is not identifiable; and a delivery component configured to provide the generated synthetic longitudinal dataset to the user or user computing device.
12. The system of claim 11, wherein the non-synthetic identifiable data comprises identifiable patient health data.
13. The system of any of claims 11-12, wherein comparing a generated synthetic longitudinal data point to the respective longitudinal data point utilized to generate the synthetic longitudinal data point comprises calculating a distance between the generated synthetic longitudinal data and the respective longitudinal data point and / or wherein each synthetic longitudinal data point is generated from a respective longitudinal data point in the plurality of longitudinal data points using a predetermined noise vector and / or wherein the generativeadversarial network utilizes a reconstruction loss function, an unsupervised loss function, and a supervised loss function when generating the synthetic longitudinal data.
14. A method for providing anonymous synthetic longitudinal data to a user, comprising the steps of: receiving, by means of a user interface (240), a query from a user or user computing device requesting a synthetic longitudinal dataset; generating, by means of a trained generative adversarial network (GAN) (263), the requested synthetic longitudinal dataset, wherein the trained GAN comprises a privacy loss function, comparing, by means of the trained GAN, a generated synthetic longitudinal data point to a respective identifiable non-synthetic longitudinal data point utilized to generate the synthetic longitudinal data point, minimizing, by means of the trained GAN, the loss function such that the generated synthetic longitudinal data point is not identifiable; and providing, by means of a delivery component, the generated synthetic longitudinal dataset to the user or user computing device.
15. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of the method of any of claims 1-5 and / or 14.
Citation Information
Patent Citations
Generative adversarial networks for synthetic data generation satisfying differential privacy
EP3910556A1
