EP4662579A1

EP4662579A1 - System and method for data encryption and decryption

Info

Publication number: EP4662579A1
Application number: EP23711987.0A
Authority: EP
Inventors: Jaiyesh CHAHAR; Indrajit Kar; Sudipta Mukhopadhyay; Vishal Pandey
Original assignee: Siemens AG; Siemens Corp
Current assignee: Siemens AG; Siemens Corp
Priority date: 2023-03-14
Filing date: 2023-03-14
Publication date: 2025-12-17
Also published as: WO2024188453A1

Abstract

A system and method for data encryption and decryption is disclosed. The method comprising encrypting original dataset using an encryption formula; transforming the encrypted data for visualization to image data; receiving the transformed encrypted data as a training data for a machine learning model for predictive analytics; and transforming the predicted results to feed them to the machine learning model for generating decrypted original dataset.

Description

Siemens Aktiengesellschaft SIE 17.178-PC 1 SYSTEM AND METHOD FOR DATA ENCRYPTION AND DECRYPTION DESCRIPTION The present invention relates to data privacy protection, and in particular to managing and controlling data privacy and security by data encryption decryption system and method en- abled by machine learning modelling. In recent times, there has been an alarming concern over individual data protection and privacy. Globally, the data privacy standards are very comprehensive, stringent and greatly restrictive in collection, retention, and use of an individual's personal data. As these data privacy laws are getting widely adopted by various jurisdictions, high penalties and fees are being levied against those who fail to comply with these regulations, ensuring data privacy compliance has become increasingly important for anyone interested and engaged in data collection, processing or usage. Thus, the primary challenge witnessed by various stakeholders involved in data exchange and transactions is to maintain a trade-off between the data privacy and data utility in an optimized manner. In the background of foregoing limitations, there exists a need for data encryption where users can perform computation on the received encrypted data from the client without first having to decrypt it. Traditional research on preserving privacy in data mining concentrates on time-invariant issues. With the rise of time series data mining, time dimensions are needed to be consid- ered for data mining techniques to uphold privacy compliances. Time-series data collected directly or indirectly by any IoT devices contain sensitive confidential information about the client that may have a negative impact on the individuals or entities whose sensitive per- sonal data is collected. This also includes selling, renting, releasing, disclosing, disseminat- ing, making available, transferring or otherwise communicating an individual’s sensitive per- sonal information to a third party without his prior authorization. Some typical examples of time-series data which carry sensitive personal information in- clude account balance, biomedical measurements, medical history, smart meter readings, life insurance policy etc. For instance, a malicious third party can access an individual’s financial statement, etc. using the shared information. Due to such privacy concerns, there needs to be devised a system and method that does not require original data from individ- uals for processing and further computation. Siemens Aktiengesellschaft SIE 17.178-PC 2 A large body of research has focused on protecting the privacy of data using different pri- vacy measures like Differential Privacy (DP), k-anonymity and information theoretic metrics. Differential Privacy prevents the service providers from inferring the current sensitive data of the user, even if the service provider has the knowledge of all the remaining private data points. Further, K-anonymity ensures that a sensitive data is indistinguishable from at least k − 1 other data points. However, DP and k-anonymity are meant to ensure the privacy of a single data point in time. Group-DP tackles this issue by applying DP for each user; how- ever, keeping many points private causes high utility losses. Additionally, these are also not appropriate measures for location trace privacy since temporal correlations are not consid- ered in certain data types such as time series data. The most established method to prevent privacy leakage for time series data is data pertur- bation technique that works only on the perturbed time-series instead of the sensitive origi- nal time-series. Only perturbed time series datasets are provided to service providers from multiple independent data sources and a data analyzer discovers meaningful mining results from the perturbed data. However, the privacy of original data can be still revealed to third party or attackers by analyzing the mining results, even if the published data is encrypted. To prevent this problem, the degree of perturbation should be made rigidly strong. In gen- eral, privacy preservation and mining accuracy are controlled by the degree of perturbation, so perturbation techniques are mainly evaluated by these two metrics, namely the degree of the privacy preservation and the loss of information. Another most popular working method dealing with privacy issue is the private data aggre- gation system. The user data is contributed to a data aggregator without compromising the user's privacy. The system can detect a private key for a local user in a set of users, wherein the sum of the private keys associated with the set of users and the data aggregator is equal to zero. The system also selects a set of data values associated with the local user. Then, the system encrypts individual data values in the set based in part on the private key to produce a set of encrypted data values, thereby allowing the data aggregator to decrypt an aggregate value across the set of users without decrypting individual data values associated with the set of users, and without interacting with the set of users while decrypting the ag- gregate value. The system also sends the set of encrypted data values to the data aggre- gator. However, private data aggregator protocols face limitations and can only compute basic functions, such as sum, average, etc. on encrypted data. In the light of above limitations, Siemens Aktiengesellschaft SIE 17.178-PC 3 there exists a need for a robust and reliable system and method for handling privacy-sen- sitive information, whereby the user is allowed to perform computation on encrypted data without having to first decrypt it. The object of the present invention is achieved by computer implemented method of data encryption-decryption, implemented upon a data privacy management system. The method comprises of encrypting original dataset using an encryption formula. This is followed by receiving the encrypted data as a training data for a machine learning model, wherein the machine learning model is trained to perform predictive analytics based on the received training data. Finally, decrypted original dataset is generated by transforming predicted re- sults obtained from the predictive analytics and feeding the transformed predicted results to the machine learning model. In one aspect, the original dataset comprises of sensor data containing one or more sensor data points that may be obtained from one or more IoT devices. In one another aspect, the original dataset is encoded into one or more images using Mar- kov transition field, Gramian angular summation field or Gramian angular difference field or a combination thereof. The encoded dataset is later transformed back to original unen- crypted dataset using computer vision techniques. In a significant aspect, the encrypted data is obtained in form of recurrence plot that is transformed back to original unencrypted dataset using unconventional approach designed for the purposes of present invention. In still another aspect, the encryption formula is derived from multiplicative and additive scaling factors followed by obtaining an encrypted data matrix for the original dataset based on the derived encryption formula. In another significant aspect, the encrypted dataset is transformed to an array, plot or ma- trix, followed by splitting the array, plot or matric into one or more image in such a way that image gets divided to vectors with position encoding to form the image data. In yet another aspect, the predicted results are fed to the machine learning model followed by merging data from obtained output of the machine learning model for generating the original decrypted dataset. Siemens Aktiengesellschaft SIE 17.178-PC 4 In still other aspect, the encrypted data is transformed to an array which is followed by splitting the array along principal diagnosis and converting to vectors with position encoding to form original dataset. In another important aspect, the predicted output is processed by descaling factors to obtain decrypted original dataset using inverse transformation. Here, the descaling factors may comprise multiplicative scaling factor, additive scaling factor, genesis value or a combina- tion thereof. In another object of the present invention, a system of data privacy management system (1000) employing data encryption-decryption is presented. The system comprising: a processing unit (500) and a memory (600) coupled to the processing unit (500). The memory (600) comprises instructions which, when executed by the processing unit (500), configures the processing unit (500) to encrypt original dataset using an encryption formula; receive the encrypted data as a training data for a machine learning model, wherein the machine leaning model is trained to perform predictive analytics based on the received training data; and generate the decrypted original dataset by transforming the predicted results obtained from predictive analytics and feeding the transformed predicted results to the machine learning model. Please note that the reference numerals 1000, 500, 600 for data privacy management system, processing unit and memory respectively, are provided for specificity and clarity, and not being shown in drawings. In still other object of the present invention, a computer-program product, having machine- readable instructions stored therein, that when executed by a processing unit (500); cause the processing unit to perform the aforementioned method steps. In yet another object of the present invention, a computer-readable storage medium com- prising instructions which, when executed by one or more processing units (500), cause the one or more processing units (500) to perform the aforementioned method steps. The present invention is not limited to a particular computer system platform, processing unit (500), operating system, or network. One or more aspects of the present invention may be distributed among one or more computer systems, for example, servers configured to provide one or more services to one or more client computers, or to perform a complete task in a distributed system. For example, one or more aspects of the present invention may be performed on a client-server system that comprises components distributed among one or more server systems that perform multiple functions according to various embodiments. Siemens Aktiengesellschaft SIE 17.178-PC 5 These components comprise, for example, executable, intermediate, or interpreted code, which communicate over a network using a communication protocol. The present invention is not limited to be executable on any particular system or group of systems, and is not limited to any particular distributed architecture, network, or communication protocol. The above-mentioned and other features of the invention will now be addressed with reference to the accompanying drawings of the present invention. The illustrated embodiments are intended to illustrate, but not limit the invention. The present invention is further described hereinafter with reference to illustrated embodiments shown in the accompanying drawings, in which: FIG.1A is a depiction of data encryption-decryption method by present method and tradi- tional method, according to an embodiment of the present invention; FIG.1B illustrates a block diagram of a data privacy system for processing data encryption- decryption, according to an embodiment of the present invention; FIG.2 depicts results of using Gramian angular field for encoding data into images; FIG.3 is an illustration of prediction from encrypted data in dense method, accord- ing to an embodiment of the present invention; FIG.4 depicts a reversal process to obtain original data from predicted results in dense method, according to an embodiment of the present invention; FIG.5 is an illustration of prediction from encrypted data in sparse method, accord- ing to an embodiment of the present invention; FIG.6 depicts a reversal process to obtain original data from predicted results in sparse method, according to an embodiment of the present invention; and Hereinafter, embodiments for carrying out the present invention are described in detail. The various embodiments are described with reference to the drawings, wherein like reference numerals are used to refer to like elements throughout. In the following description, for pur- pose of explanation, numerous specific details are set forth in order to provide a thorough Siemens Aktiengesellschaft SIE 17.178-PC 6 understanding of one or more embodiments. It may be evident that such embodiments may be practiced without these specific details. The embodiments herein provide steps that may be performed in a computer system such as a set of computer-executable instructions and, although a logical order is illustrated in the flow charts, in some cases, the steps illustrated or described may be performed in an order different than here. Before beginning a more detailed discussion of the various aspects of the illustrative embodiments, it should first be appreciated that throughout this description the term “method” or “process” will be used to refer to aspects of the present disclosure that perform various operations, functions, and the like. A “process,” as the term is used herein, may be an implementation of the functions or aspects of the illustrative embodiments in the form of a system, a method, or a computer program product. In the case of a method, the method steps are implemented by one or more devices, apparatus, computers, data processing systems, or the like. In the case of a computer program product, the logic represented by computer code or instructions embodied in or on the computer program product is executed by one or more hardware devices in order to implement the functionality or perform the operations associated with the specific “process.” Thus, the process described herein may be implemented as specialized hardware, software executing on general-purpose hardware, software instructions stored on a medium such that the instructions are readily executable by specialized or general purpose hardware, a procedure or method for executing the functions, or a combination of any of the above. In addition, it should be appreciated that the following description uses a plurality of various examples for various elements of the illustrative embodiments to further illustrate example implementations of the illustrative embodiments and to aid in the understanding of the mechanisms of the illustrative embodiments. These examples intended to be non-limiting and are not exhaustive of the various possibilities for implementing the mechanisms of the illustrative embodiments. It will be apparent to those of ordinary skill in the art in view of the present description that there are many other alternative implementations for these various elements that may be utilized in addition to, or in replacement of, the examples provided herein without departing from the spirit and scope of the present invention. In one general embodiment of present disclosure, a data encryption-decryption system and method (1000) is proposed where users can perform computation on the received encrypted data from the client without having to first decrypt it. This allows the data to be Siemens Aktiengesellschaft SIE 17.178-PC 7 accessible in commercial cloud environments for processing and further computation, enabling more efficient use of the data while meeting requirements for user privacy protection, data security, government regulations, and the like. This also improves the processing efficiency of training a prediction model when the data volume is large, and simultaneously ensure the security and privacy of private data. As commonly known, visualization of data is a graphical method designed to locate hidden patterns, outliers that defy those patterns, non-stationarity, correlations, trends, seasonality and structural changes in any data type. Here, for exemplary purposes reference is made to time series data or temporal data obtained from one or more sensors. In accordance with one example embodiment, this nonlinear dynamic tool applied to the study of data, say time series data and is designed to identify non-obvious dependencies in the time series data dynamics. Accordingly, in the presently proposed solution, the encrypted data is processed directly with analytical functions to yield same encrypted results as when these analytical functions would have been made to run on original unencrypted data. Moreover, with the present approach, any kind of mathematical operation can be performed for an unlimited number of times. Referring now to Fig. 1A, data encryption-decryption method by machine learning model using quasi recurrence plot analysis is presented as an example embodiment. In one advantageous embodiment, using quasi-recurrent neural network better accuracy and high throughput is achieved in predicting future time series values. As can be seen in Fig.1A, in conventional method the original observations or dataset (10) are fed to the machine learning model (50) to obtain results (12), which eventually led to privacy revelation (13). On the contrary, with the original observations or dataset (10) is first encrypted to obtain encrypted dataset (20), followed by feeding the encrypted dataset (20) to the machine learning model (50), the results obtained (18) do not reveal any aspects of original dataset (10) and hence privacy preserved results are obtained (19). Referring to Fig. 1B, overall working procedure for present disclosure is described. Accordingly, the method adopted for data encryption-decryption involves obtaining sensitive data from the client (100) followed by encrypting original dataset (client sensitive data) using an encryption formula to obtain encrypted dataset (20). This is followed by transforming the encrypted dataset (20) for visualization to image data. Now the transformed encrypted data is received by a machine learning model (50) for predictive analytics, wherein the machine leaning model (50) is trained to perform predictive analytics based on the received training Siemens Aktiengesellschaft SIE 17.178-PC 8 data. Finally, the predicted result (80) is transformed and fed to machine learning model (50) for generating decrypted original dataset (10). Re-referring to Fig.1A, by adopting the traditional method there are possibilities of privacy revelation (70) when the original dataset (10) is fed directly to the machine learning model (50). On the contrary, in the proposed disclosure, the encrypted dataset (20) is obtained from client (100) in the form of recurrence plot or array. As an extension of the recurrence plots, the Recurrence Quantification Analysis (RQA) defines measures for diagonal segments in a recurrence plot, recurrence rate, determinism, averaged length of diagonal structures, entropy, and trend due to the preservation of all primary information along the principal diagonal. As understood, the recurrence rate is the ratio of all recurrent states (recurrence points) to all possible states and is the probability of recurrence of a special state. Further, stochastic behavior causes very short diagonals, whereas deterministic behavior causes longer diagonals. Therefore, the ratio of recurrence points forming diagonal structures to all recurrence points is a measure of the determinism in the system. Diagonal structures show the range in which a piece of the trajectory is rather close to another piece of the trajectory at different times. The diagonal length is the time how long they will be close to each other and can be interpreted as the mean prediction time. The entropy means the Shannon entropy of the histogram of line segment lengths and reflects the complexity of the deterministic structure in the system. While on one hand the stationary systems deliver homogeneous recurrence plots, on the other hand the non-stationary systems cause changes in the distribution of recurrence points in the plot. For example, a simple drift causes a brightening of the recurrence plot away from the main diagonal to the edges. The value trend measures this effect by diagonal-wise computation of diagonal recurrence density and its linear relation to the time distance of these diagonals to the main diagonal. In one example embodiment, Gramian Angular Summation Field and Gramian Angular difference field is used for encoding the time series data into images, and these images can then be used for solving time series problems using computer vision. Here, instead of normal cartesian coordinates, time series data is represented in polar coordinates while using gramian angular field matrix for encoding. Each element of the matrix is cosine of summation of angles between temporal points. Siemens Aktiengesellschaft SIE 17.178-PC 9 For a time series T = [t1,t2,……,tm] having m points, first of all T is rescaled to interval -1 to 1 or 0 to 1: The time series after rescaling is represented in polar coordinates by encoding the value of elements as cosine with time stamps in radians. ^^ = arccos( ^^ _^^̃) − 1, −1 ≤ ^^ _^^ ≤ 1̃, ^^ _^^̃ ∈ ^^ ^(11.3) Where, Xi is time stamp with M as a constant factor for regularizing polar coordinate system span. With increase in time, values warp at different angular points of the spanning circles. As cosine is used, so the encoded series will be bijective for angle Φ between 0 to π. So, in polar coordinate system, the time series will produce one result and also with inverse function. And instead of cartesian coordinates, polar coordinates are used, so the temporal information of the time series data is also preserved. Next step after the conversion to polar coordinate system is to take the trigonometric sum of pair of points for calculating the temporal correlation at different time points. Gramian summation angular field is represented as: = _{^^ ^^̃.} ^(11.5) Siemens Aktiengesellschaft SIE 17.178-PC 10 Gramian Difference angular field (GADF) is represented as: [^{sin ( ^^2 + ^^1) ⋯ sin ( ^^2 + ^^ ^^)} = _{] (11.6)} GADF _{⋮ ⋮} _{sin ( ^^ ^^ + ^^2) ⋯ sin ( ^^ ^^ + ^^ ^^)} = I is unit vector [1,1,1,…..,1]. Advantages of using GAFs: By using GAFs, the temporal dependency between data points is preserved. From the main diagonal of matrices, time series can be reconstructed with high level features learned by neural networks. Different patterns of time series are enhanced by GAFS for visual inspection of humans. The complete illustration of conversion to GAF is shown in Fig.2. In other alternate embodiment, Markov transition field is used, which is transition field to represent the Markov transition probabilities sequentially to preserve the information in the time domain. The Markov transition field generation process aligns each probability along the temporal order to build the Markov transition field for the signal. For a time-series T, the quantile bins are represented by Q and each of the tj are assigned to their corresponding bins qk (where k lies between 1 and Q). Using this an adjacency matrix A can be formulated by counting transitions amongst quantile bins with respect to first order Markov chain along the time axis. Thus, the derived Markov Transition Matrix A is: where ajk is the frequency with which a point in quantile qj follows a point in quantile qk and is normalized as Σk ajk = 1. This matrix does not consider the proper distribution of time-series T and temporal dependency on the time-steps sj. Therefore, to get rid of this temporal dependency without losing too much information, the Markov Transition Field is defined as: Siemens Aktiengesellschaft SIE 17.178-PC 11 Hence, a Markov transition matrix is formed by dividing the data into Q quantile bins, where the data at time-steps j and k are contained with the bins qj and qk. At any point the transition probability of qj qk in the Markov Transition Field is given by Mjk. Assigning this probability from quantile at time-step j to k, the MTF M helps in encoding multi-step transition probabilities of the time-series. Now, re-referring to encrypted data that is obtained in form of recurrence plot, the above proposed recurrence plot data is encrypted on following using an encryption formula for which the basis function is as described below: def enc (s, eps, z): m=np. zeros((len(s), len(s))) for i in range (len(s)): m[i]=s[i]-s+z mf=m/eps return mf where s is the original time series; eps and z are the multiplicative and additive scaling factor respectively. mf is the encrypted data matrix. This encrypted data (mf) (20) is visualized in the form of image, which is sent by client (100) to the service providers (200). In next working embodiment, the image representative of encrypted data (20) is fed to a machine learning model (50), where the data is analysed to map input data to output data according to a machine-learning algorithm. The machine-learning model is trained on above image data that provides interactive visual analytics and enables forecasting or finding solutions for varied scenarios such as classification, regression, anomaly, finding future values in the form of image. However, this data is not required to be decrypted from the given image for further computation and forecasting. To obviate the need for decrypting the data, the present disclosure proposes following methods: Siemens Aktiengesellschaft SIE 17.178-PC 12 a) Dense Method b) Sparse Method c) Split Method Next, in accordance with one working embodiment of present disclosure, each of these methods will be discussed in sufficient detail for better understanding and clarity. a) Dense Method: Referring to Fig.3, the data or original dataset (10) received as an input is first converted to a plot (15), which is then split into image patches (17) in such a way that the image gets divided to vectors with position encoding. These training set patches will now serve as an input to any of existing AI/ML models (50). In one example embodiment, this AI/ML model may be a generative autoencoder model (though not limited to autoencoders), which gets compiled and transforms data from a higher dimensional space to a lower dimensional space. In one exemplary embodiment, self-adaptive adadelta optimization is implemented that continues to learn throughout the training without relying on initial learning rate and shows no drop in performance because of continual decay of learning rates. Further, mean squared error is used as a loss function that helps better reconstruct the input with autoencoders. Finally, the model is fit on training set and executed to predict from validation set patches with a high accuracy rate. In one example embodiment, the predicted results or the output (18) are transformed for future value into images. In a reversal process, as shown in Fig. 4, the predicted image or output (18) is fed back into the autoencoder model (50) and from the obtained output, the image patches are descaled (21) and merged (22) to retrieve the original recurrence plot or array (15). b) Sparse Method: In this method, even the whole recurrence plot (15) after encryption is not required. Rather, the plot (15) of encrypted data (20) is split only along the principal diagonal (24) and then convert (25) to vectors with position encoding, as shown in Fig. 5. The diagonal information is thus sufficient, only patches along the principal diagonal (24) of the recurrence plot (15) is required and the client (100) can perform recurrence quantification analysis. The next sequence of steps in this method are same as Dense Method. Siemens Aktiengesellschaft SIE 17.178-PC 13 The diagonals are only used to train AI models (50). The model is fit with best suited optimizer on the training set and thus prediction can be made from the validation set patches with a high accuracy rate and the least error. The scaling factor /basis function and non-diagonal data is needed to build the original dataset (10), (time series data in this example scenario) and the methodology can be applied to any time dependent data. The predicted results or output (18) are transformed for future value into the image. Likewise, as shown in Fig.6, reverse of this process is achieved in a manner discussed for dense method above. Accordingly, in the reverse way, the predicted image (18) is fed into the training model (50) and after getting the result, the patches are descaled (21) and merged (22) to get back the diagonal-wise principal information of the recurrence plot (15). Using the inverse transformation, the client can descale the output (18) and return the original dataset (10). Hence, from the privacy perspective, this is a double-encrypted data. The training set patches here work as input in the machine-learning model (50). Finally, the model (50) is fit with best suited optimizer on the training set and can predict from the validation set patches with high accuracy rate and least error. The predicted results (18) are transformed for future value into the image. In the reverse way, the predicted image (18) is fed into the training model (50) and after getting the result, the patches are merged to get back the diagonal-wise principal information of the recurrence plot (15). Thus, using the inverse transformation the client (100) can descale the output and return the original dataset (10). The mechanism of encoding the input dataset (10) is by transforming it into another liner space by scaling (14). The keys used for this transformation are the slope and intercept values and hereby termed as the genesis values or keys. This genesis value or the keys are shared among all the parties involved (i.e. client and the service provide for in- stance). The encoded data (20) is then further fed into subsequent stages. The inverse transformation is again done by performing the inverse operations using the same gen- esis values or keys shared between the both the parties. Both of the steps discussed above are pretty straightforward and doesn't require heavy or complex computations. c) Split Method: The times series data is scaled/basis function is applied to extract respective plot types. While the 1st two types of plot hold the static information of time series by exploring the angular properties using trigonometry, the last holds dynamic Siemens Aktiengesellschaft SIE 17.178-PC 14 aspects of time series. Thereafter the plots/images are split into image patches in a way such that the image will be divided to vectors with position encoding. Lastly, to retrieve the descaled output, the client (100) runs the predicted data (18) obtained as an output from any AI based algorithm. This predicted data (18) is fed to the AI model (50) of present disclosure that has learned to encode an image and generate decoded original dataset (time series) (10). This time series decrypted data is then passed to descaling factors such as-multiplicative scaling factor (eps), additive scaling factor (z) and genesis value (t0). Here, ret is the returned array containing all values of original time series data. In one example embodiment, the proposed decryption formula applied is: def dec (mf, eps, t0, z): arr=s*eps mar=arr [0]-z ret=t0-mar return ret Thus, with the proposed data encryption-decryption method of present disclosure, one can leverage cloud computing and storage services securely. It optimally balances the trade-off between data security and usability. Further, the method enables organizations to share sensitive business data with third parties without revealing the data or the results of the computation to them. This can accelerate collaboration and innovation without the risk of sensitive information getting compromised. Furthermore, the method enables business businesses operating in heavily regulated industries, such as healthcare and finance, to get outsourcing services for research and analytical purposes without the risk of non-compli- ance. Thus, with the proposed method and system for data encryption-decryption, privacy and utility of data that contain information of a sensitive or identifying nature can be managed and controlled. Those of ordinary skill in the art will appreciate that the chosen hardware may vary depend- ing on the implementation. Other internal hardware or peripheral devices, such as flash memory, equivalent non-volatile memory, or optical disk drives and the like, may be used in addition to the above mentioned hardware. Also, the processes of the illustrative embod- iments may be applied to a multiprocessor data processing system, without departing from the spirit and scope of the present invention. Siemens Aktiengesellschaft SIE 17.178-PC 15 The present invention can take a form of a computer program product comprising program modules accessible from computer-usable or computer-readable medium storing program code for use by or in connection with one or more computers, processors, or instruction execution system. For the purpose of this description, a computer-usable or computer-read- able medium can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The medium can be electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system (or apparatus or device) or a propagation mediums in and of themselves as signal carriers are not included in the definition of physical computer- readable medium include a semiconductor or solid state memory, magnetic tape, a remov- able computer diskette, random access memory (RAM), a read only memory (ROM), a rigid magnetic disk and optical disk such as compact disk read-only memory (CD-ROM), com- pact disk read/write, and DVD. Both processors and program code for implementing each aspect of the technology can be centralized or distributed (or a combination thereof) as known to those skilled in the art. The foregoing examples have been provided merely for the purpose of explanation and are in no way to be construed as limiting of the present invention disclosed herein. While the invention has been described with reference to various embodiments, it is understood that the words, which have been used herein, are words of description and illustration, rather than words of limitation. Further, although the invention has been described herein with reference to particular means, materials, and embodiments, the invention is not intended to be limited to the particulars disclosed herein; rather, the invention extends to all functionally equivalent structures, methods and uses, such as are within the scope of the appended claims.

Siemens Aktiengesellschaft SIE 17.178-PC 16 LIST OF REFERENCE NUMBERS: Observations (Decrypted results): 10 Result from conventional method: 12 Privacy revealed: 13 Scale: 14 Plot/Array: 15 Split: 17 Predicted result of present application: 18 Privacy preserved step: 19 Encrypted Data: 20 Descale: 21 Merge: 22 Extract diagonal: 24 Position encoded vectors: 25 Machine learning model: 50 Client: 100 Processing Unit: 500 Memory: 600 Data Privacy Management System: 1000

Claims

Siemens Aktiengesellschaft SIE 17.178-PC 17 CLAIMS We Claim: 1. A computer implemented method of data encryption-decryption, implemented upon a data privacy management system (1000), the method comprising: encrypting, by a processing unit (500), original dataset (10) using an encryption formula; receiving, by the processing unit (500), the encrypted dataset (20) as a training data for a machine learning model (50), wherein the machine leaning model (50) is trained to perform predictive analytics based on the received training data; and generating, by the processing unit (500), decrypted original dataset (10) by transforming predicted results obtained from the predictive analytics and feeding the transformed predicted results (18) to the machine learning model (50). 2. The method according to claim 1, wherein the original dataset (10) comprises of sensor data containing one or more sensor data points. 3. The method according to claim 1, wherein the method of encrypting the original dataset (10) comprises of encoding the original dataset (10) into one or more images using Markov transition field, Gramian angular summation field or Gramian angular difference field or a combination thereof. 4. The method according to claim 3, further comprising transforming the encrypted dataset (20) to the original dataset (10) using computer vision techniques. 5. The method according to claim 1, wherein the method of encrypting the original dataset (10) comprises of obtaining the encrypted dataset (20) in form of recurrence plot (15), and transforming the encrypted dataset (20) to the original dataset (10). 6. The method according to claim 1, wherein the method of encrypting the original dataset (10) using the encryption formula comprises of: deriving the encryption formula from multiplicative and additive scaling factors (14); and obtaining an encrypted data matrix for the original dataset (10) based on the derived encryption formula. Siemens Aktiengesellschaft SIE 17.178-PC 18 7. The method according to claim 1, further comprising: transforming the encrypted dataset (20) to an array, plot or matrix (15); and splitting (17) the array, plot or matrix (15) into one or more images in such a way that image gets divided to vectors with position encoding to form image data. 8. The method according to claim 1, wherein the method of generating the original de- crypted dataset (10) comprises of: feeding the predicted results (18) to the machine learning model (50); and merging data (22) from obtained output of the machine learning model(50). 9. The method according to claim 1, further comprising: transforming the encrypted data (20) to an array (15); and splitting the array (15) along principal diagonal (24) and converting to vectors (25) with position encoding to form the original dataset (10). 10. The method according to claim 1, wherein method of generating the decrypted original dataset (10) comprises of processing the predicted results (18) by descaling factors (21) and using inverse transformation. 11. The method according to claim 10, wherein the descaling factors (21) comprise multi- plicative scaling factor, additive scaling factor, genesis value or a combination thereof. 12. A data privacy management system (1000) employing data encryption-decryption, comprising: a processing unit (500), and a memory (600) coupled to the processing unit (500), wherein the memory (600) comprises instructions which, when executed by the processing unit (500), config- ures the processing unit (500) to perform the method steps according to any of the claims 1 to 11. 13. A computer-program product, having machine-readable instructions stored therein, that when executed by a processor, cause the processing unit to perform method steps according to any of the claims 1-11. Siemens Aktiengesellschaft SIE 17.178-PC 19 14. A computer-readable storage medium comprising instructions which, when executed by one or more processing units (500), cause the one or more processing units (500) to perform a method according to any of the claims 1-11