A cloud platform and method for efficiently processing aggregated data

Abstract

A computer implementation method for efficiently processing aggregated data shared by users of a cloud platform (1), the method comprising the steps of: uploading (S1) at least one dataset (DS) to the cloud platform (1) by a user's client device; calculating (S2) a similarity score indicating the degree of similarity between the currently uploaded dataset (DS) and a dataset (DS) previously uploaded by another user's client device; and performing (S3) a user selection process on the cloud platform (1) based on the aggregated data, the aggregated data including the current dataset (DS) of the corresponding user and a dataset (DS) previously uploaded from another user's client device and stored in a database (3) of the cloud platform (1), the previously uploaded dataset (DS) having a calculated similarity score that exceeds a configurable similarity score threshold related to the currently uploaded dataset (DS) of the corresponding user.

Description

[0001] This invention relates to a computer implementation method for efficiently processing pooled data shared by users of a cloud platform.

[0002] A cloud platform connects to multiple client devices belonging to different users or clients via a network. These client devices can upload data to the cloud platform's database. This data can include, for example, sensor data generated by sensors on machines in the respective user's industrial system. Furthermore, the cloud platform can provide various services to its users. These services can include processes executed on user data stored in the cloud platform's database. For example, a user can invoke an analytics process on the cloud platform that analyzes stored user data (such as sensor data) to discover useful information. The platform returns the results of the analytics process to the user's client device via the network. These processes can also include training, testing, and / or inference processes for machine learning used in data models. However, conventional cloud platforms do not provide a mechanism that allows users to automatically aggregate their individual data to generate a data pool, enabling efficient processing of the aggregated data within the cloud platform.

[0003] US 2018 / 181641 A1 discloses a method for recommending analytical tasks based on the similarity of datasets. A system includes a data processor, a matching module, and a recommendation module. The data processor receives an incoming dataset and generates feature vectors for the incoming dataset. The matching module determines a similarity measure between the generated feature vectors and representative feature vectors from a plurality of datasets in a database, and selects at least one dataset from the plurality of datasets based on the similarity measure. The recommendation module identifies at least one analytical task associated with the selected dataset and recommends at least one analytical task to be performed on the incoming dataset.

[0004] US 2014 / 195818 A1 discloses a user device that encrypts data and privacy attributes associated with that data. The processing device receives the encrypted data and privacy attributes, receives a signed script from a requester, and verifies the signature. If verification is successful, a private key is unsealed and used to decrypt the privacy attributes and script attributes, which are compared to determine if the script respects the privacy attributes. If so, the encrypted data is decrypted, and the script processes the private data to generate a result encrypted using the requester's key, which is then output.

[0005] Therefore, the object of the present invention is to provide a method and system for efficiently processing aggregated data shared by users of a cloud platform.

[0006] According to a first aspect of the invention, this objective is achieved by a computer implementation method comprising the features of claim 1.

[0007] According to a first aspect, the present invention provides a computer implementation method for efficiently processing aggregated data shared by users of a cloud platform, the method comprising the following steps:

[0008] The user's client device uploads at least one dataset to the cloud platform.

[0009] A similarity score is calculated, indicating the degree of similarity between the currently uploaded dataset and datasets previously uploaded by other users' client devices.

[0010] The process of user selection on the cloud platform is performed based on aggregated data, which includes the corresponding user's current dataset and datasets previously uploaded from other users' client devices and stored in the cloud platform's database. The datasets previously uploaded from other users' client devices have a calculated similarity score that exceeds a configurable similarity score threshold relative to the corresponding user's currently uploaded dataset.

[0011] Client devices of users who have already uploaded the current dataset to the cloud platform receive recommendation messages from the cloud platform to aggregate datasets from other users on the cloud platform that match the current dataset.

[0012] If the cloud platform receives an acceptance message from the user's client device to aggregate the dataset, it will automatically aggregate the matching datasets to generate a dataset pool.

[0013] In a possible embodiment of the method according to the first aspect of the invention, the process performed by the cloud platform based on the aggregated data includes a training process for training a data model.

[0014] In a further possible embodiment of the method according to the first aspect of the invention, the process performed by the cloud platform based on the aggregated data includes a testing process for testing the trained data model.

[0015] In a further possible embodiment of the method according to the first aspect of the invention, the process performed by the cloud platform based on the aggregated data includes an inference process for performing a trained and tested data model.

[0016] In a possible embodiment of the method according to the first aspect of the invention, the data model used includes an artificial neural network.

[0017] In a further possible embodiment of the method according to the first aspect of the invention, for each uploaded dataset, a representation vector is computed, the representation vector comprising vector elements representing the statistical properties of the uploaded dataset.

[0018] In a further possible embodiment of the method according to the first aspect of the invention, a similarity score indicating the degree of similarity between the uploaded datasets is calculated based on the representation vector of the uploaded dataset.

[0019] In a further possible embodiment of the method according to the first aspect of the invention, the calculated similarity score includes a cosine similarity score.

[0020] In a further possible embodiment of the method according to the first aspect of the invention, if the similarity score calculated for the currently uploaded dataset and related to a previously uploaded dataset of another user exceeds a configurable similarity score threshold, the corresponding previously uploaded and stored dataset is marked as a matching dataset with respect to the dataset currently uploaded by the user's client device.

[0021] In a further possible embodiment of the method according to the first aspect of the invention, the aggregated data includes datasets uploaded from client devices of different users and labeled as matching datasets.

[0022] In a further possible embodiment of the method according to the first aspect of the invention, a similarity score is triggered to calculate the similarity scores of previously uploaded datasets of other users stored in a database of the cloud platform in response to the upload of a new current dataset from the client device of the respective user.

[0023] In a further possible embodiment of the method according to the first aspect of the invention, the matching datasets undergo homomorphic encryption before they are aggregated to generate a dataset pool.

[0024] In a further possible embodiment of the method according to the first aspect of the invention, a user-selected process is performed on a cloud platform to calculate a benchmark based both on a pool of generated datasets and on the current dataset uploaded to the cloud platform by the user's client device. The benchmark indicates an increase in efficiency in processing the uploaded dataset through the selected process, the increase in efficiency being caused by data aggregation.

[0025] In a further possible embodiment of the method according to the first aspect of the invention, the calculated benchmark is sent by the cloud platform to the user's client device.

[0026] In a possible embodiment of the method according to the first aspect of the invention, the dataset includes labeled data.

[0027] In a further possible embodiment of the method according to the first aspect of the invention, the dataset includes unlabeled data.

[0028] According to a further second aspect, the present invention further provides a cloud platform including the features of claim 16.

[0029] According to a second aspect, the present invention provides a cloud platform for efficiently processing aggregated data shared by users of a cloud platform, wherein the cloud platform includes:

[0030] The database is adapted to store datasets uploaded to the cloud platform by users' client devices.

[0031] A scoring unit is configured to calculate similarity scores, each score indicating the degree of similarity between the currently uploaded dataset and previously uploaded datasets stored in the database.

[0032] A processor adapted to execute a user-selected process based on aggregated data, which includes the current dataset uploaded by the corresponding user and previously uploaded datasets from other users, the previously uploaded datasets having calculated similarity scores that exceed a configurable similarity threshold relative to the current dataset uploaded by the corresponding user.

[0033] Client devices of users who have already uploaded the current dataset to the cloud platform receive recommendation messages from the cloud platform to aggregate datasets from other users on the cloud platform that match the current dataset.

[0034] If the cloud platform receives an acceptance message from the user's client device to aggregate the dataset, it will automatically aggregate the matching datasets to generate a dataset pool.

[0035] In the following, different aspects of possible embodiments of the invention are described in more detail with reference to the accompanying drawings.

[0036] Figure 1 A flowchart illustrating a possible exemplary embodiment of a computer-implemented method according to a first aspect of the present invention is shown;

[0037] Figure 2 A block diagram illustrating a possible exemplary embodiment of a cloud platform according to a further aspect of the present invention is shown;

[0038] Figure 3 A signal diagram illustrating the operation of a computer implementation method according to the present invention for efficiently processing aggregated data shared by users of a cloud platform.

[0039] from Figure 1As can be seen from the flowchart, in the illustrated embodiment, the computer implementation method for efficiently processing aggregated data shared by users of cloud platform 1 according to the first aspect of the present invention includes three main steps.

[0040] In the first step S1, at least one dataset DS is uploaded to cloud platform 1 by the user's client device. The dataset DS is uploaded from the user's client device to server 2 of cloud platform 1 via the network. Server 2 also... Figure 2 The scoring calculation unit 2A and processing unit 2B are implemented as illustrated in the block diagram. The dataset DS may include labeled or unlabeled data. The dataset DS may, for example, include sensor data generated by sensors on a user's machine and uploaded to cloud platform 1 by a client device. The client device may include edge devices of the automated system. In possible embodiments, user data, such as sensor data, may undergo preprocessing performed by the processor of the client device. In possible embodiments, the dataset DS may include data structures or data groups. The data structure of the dataset DS may include metadata describing the uploaded user data. In possible embodiments, the client device pushes the dataset DS to cloud platform 1 via a network. In alternative embodiments, cloud platform 1 may, for example, periodically pull the dataset DS from the user's client device.

[0041] In a further step S2, a similarity score SS is calculated, which indicates the degree of similarity between the currently uploaded dataset DS and other datasets DS previously uploaded by other users' client devices. In a possible embodiment, for each uploaded dataset DS, a representation vector is calculated, which includes vector elements representing the statistical properties of the uploaded dataset DS. These statistical properties may include, for example, the mean or standard deviation. The representation vector may include the most common statistical features of a given dataset DS. In a possible embodiment, in step S2, a similarity score SS indicating the degree of similarity between uploaded datasets is calculated based on the representation vectors of the uploaded datasets DS. In a possible embodiment, the calculated similarity score SS may include a cosine similarity score. In a possible embodiment, if the similarity score SS calculated for the currently uploaded dataset DS and related to a previously uploaded dataset DS by another user exceeds a configurable similarity score threshold, the corresponding previously uploaded and stored dataset DS can be marked as a matching dataset DS with respect to the dataset DS currently uploaded by the user's client device. Therefore, whenever the pairwise similarity between two datasets DS exceeds a configurable threshold, platform 1 can represent them as a match.

[0042] In a further step S3, the process selected by the user on cloud platform 1 is performed based on the aggregated data. The aggregated data may include the user's current dataset DS recently uploaded by the corresponding user and datasets DS previously uploaded from other users' client devices and stored in database 3 of cloud platform 1, which have a calculated similarity score SS that exceeds the configurable similarity score threshold TH related to the corresponding user's currently uploaded dataset DS.

[0043] In a possible embodiment, the configurable similarity score threshold TH is set by the service provider of cloud platform 1. In an alternative embodiment, the configurable similarity score threshold TH can be applied through a process selected by the user to be executed by processing unit 2B of cloud platform 1. In this embodiment, the user of cloud platform 1 can adjust the desired similarity score threshold TH to define the degree to which other users' datasets DS must be similar to their own provided dataset DS.

[0044] To select the process in step S3, the user's client device can invoke a process provided by cloud platform 1. This process could be, for example, an analysis process that performs data analysis on a pooled dataset DS. The analysis process could be, for example, a predictive maintenance process that predicts when components of a user's automated system might fail. This predictive maintenance process is more accurate when executed on multiple pooled datasets DS comprising multiple users who have similar or identical machines in their respective automated systems.

[0045] In possible embodiments, the user-selected process may further include a training process for training the data model, particularly an artificial neural network (ANN). The user-selected or invoked process may also include a testing process for testing the trained data model, particularly the trained ANN. The user-selected or invoked process may further include an inference process for performing the trained and tested data model, particularly the trained and tested ANN.

[0046] Cloud platform 1 performs aggregation of datasets DS based on the calculated similarity score SS. Aggregated data includes datasets DS uploaded from client devices of different users and labeled as matching datasets DS. In a possible embodiment, the calculation of similarity scores SS for previously uploaded datasets DS by other users, stored in database 3 of cloud platform 1, can be triggered in step S2 in response to the upload of a new current dataset DS from the corresponding user's client device. The client device of the user who has uploaded the current dataset DS to cloud platform 1 receives a recommendation message from cloud platform 1 to aggregate datasets DS of other users of cloud platform 1 that match the current data—that is, have a calculated similarity score SS that exceeds a configurable similarity score threshold TH related to the corresponding user's currently uploaded dataset DS. Matching datasets DS are automatically aggregated to generate a dataset pool only if cloud platform 1 receives an acceptance message from a user's client device to aggregate datasets DS. Users have complete control over whether their datasets DS are aggregated with those of other users.

[0047] In a possible embodiment of the computer implementation method according to the invention, the matching datasets DS undergo homomorphic encryption before they are aggregated to generate dataset DS. Therefore, homomorphic encryption is applied to the data before it is shared among users. Homomorphic encryption is a method of encrypting data that allows computation to be performed such that the result of the computation, when encrypted, matches the result of the same computation process on unencrypted data. Homomorphic encryption is used for securely outsourced computation, i.e., executing selected processes on the processing resources of cloud platform 1.

[0048] In a further possible embodiment of the computer implementation method according to the first aspect of the invention, a user-selected process is executed on the processor of cloud platform 1, based both on the generated dataset pool and on the current dataset DS uploaded to cloud platform 1 by the user's client device, to compute a benchmark indicating an increase in efficiency in processing the uploaded dataset DS as the selected process, the efficiency increase being caused by data aggregation. In a possible embodiment, the computed benchmark may be sent back to the user's client device by cloud platform 1. In this embodiment, the user is notified of the impact of data aggregation on the results of the executed process. The user may be notified of the efficiency increase in executing the process caused by data aggregation.

[0049] Figure 2 A block diagram illustrating a possible exemplary embodiment of a cloud platform 1 according to one aspect of the present invention is shown. The cloud platform 1 is used for efficient processing of aggregated data shared by users of the cloud platform 1. Multiple client devices can connect to the cloud platform 1 via a network, particularly the Internet. Figure 2The cloud platform 1 illustrated may include a server 2 with access to a database 3. Database 3 may be a central database or a distributed database. In the illustrated exemplary embodiment, server 2 may implement a score calculation unit (SCU) 2A and a processing unit (PU) 2B. The score calculation unit 2A is configured to calculate similarity scores SS, each SS indicating the degree of similarity between at least one currently uploaded dataset DS and a previously uploaded dataset DS stored in the database 3 of cloud platform 1. The processing unit 2B of server 2 may be adapted to execute a process selected or invoked by a user based on aggregated data, which includes the currently uploaded dataset DS of the corresponding user and previously uploaded datasets of other users, the previously uploaded datasets of other users having calculated similarity scores SS that exceed a configurable similarity threshold TH related to the currently uploaded dataset DS of the corresponding user. The similarity threshold TH may be configured by the provider of cloud platform 1 or may be submitted to cloud platform 1 by the user's client device when invoking a corresponding process. In possible embodiments, the process executed by processing unit 2B based on the aggregated data may include an analysis process that analyzes the aggregated data. The process executed by processing unit 2B may further include a training process for training a data model, a testing process for testing the trained data model, and / or an inference process for executing the trained and tested data model. In possible embodiments, the data model may include an artificial neural network (ANN), such as a feedforward neural network or a recurrent neural network. The ANN may consist of an input layer, several hidden layers, and an output layer. In possible embodiments, the user may indicate whether the uploaded dataset DS will be used for the training, testing, and / or inference processes of a specific data model. In possible embodiments, the user may also specify the type of data model used, particularly the type of ANN used for the corresponding invoked process. Processing unit 2B is adapted to execute the user-selected process based on aggregated data, which includes the user's currently uploaded dataset DS and other users' previously uploaded datasets DS, the latter having a calculated similarity score SS that exceeds a configurable similarity threshold TH relative to the user's currently uploaded dataset DS. The calculation result of this process may be returned to the client device of the user who invoked the process. Furthermore, in a possible embodiment, the user's client device may receive a benchmark indicating an increase in efficiency in processing the uploaded dataset DS through a selected process, the increase in efficiency being caused by the performed data aggregation.

[0050] Figure 3A diagram illustrating the operation of a computer implementation method for efficiently processing aggregated data shared by users of cloud platform 1 is shown. In the illustrated example, two users A and B are connected to a public cloud platform CP via a data network, such as... Figure 2 The cloud platform 1 is illustrated in the diagram. In the example shown, the first user A will use the dataset DS. A Uploaded to the cloud platform CP. Based on the uploaded dataset DS A Cloud platform 1 calculates the representation vector V A This represents the vector V. A Includes the uploaded dataset DS A Vector elements with statistical properties (such as mean or standard deviation).

[0051] When another user B puts the dataset DS B When uploading to the cloud platform, it is used for Figure 3 The representation vector V is calculated in the same way as shown in the figure. B Based on the two calculated representation vectors V A V B A similarity score SS is calculated using a predefined similarity scoring function. In a possible embodiment, the similarity score SS calculated by the processing unit 2B of the cloud platform 1 includes a cosine similarity score. The processing unit 2B includes a comparator adapted to compare the calculated similarity score SS with a configurable similarity score threshold TH, such as... Figure 3 As illustrated in the diagram. In the illustrated embodiment, if the calculated similarity score SS exceeds a pre-configured similarity score threshold TH, users A and B both receive a recommendation REC from cloud platform 1 to share their data. Server 2 of cloud platform 1 transmits the recommendation REC to the client devices of users A and B. The client device of a user who has already uploaded the current dataset DS to cloud platform 1 receives the recommendation message REC from cloud platform 1 to aggregate their dataset DS with datasets DS from other users on cloud platform 1 that match the current dataset DS (i.e., have a similarity score SS exceeding the threshold TH). If cloud platform 1 receives an acceptance message ACC from the corresponding user's client device to aggregate the datasets DS, it automatically aggregates the matching datasets DS to generate a dataset pool, as shown in the diagram. Figure 3 As illustrated in the diagram. In a possible embodiment, the matching datasets DS automatically undergo homomorphic encryption performed by the processing unit 2B of server 2 before they are aggregated to generate a dataset pool. Based on the aggregated data, a user-selected process P is performed on cloud platform 1. Figure 3 In the example illustrated, user A calls not only to user A's dataset DS A Execute and also on user B's dataset DS BThe execution process P. The result of the called process is returned to user A's client device, such as... Figure 3 As shown in the diagram. Similarly, if another user B invokes the same procedure P, then the procedure is also executed on the aggregated data, which includes user A's dataset DS. A And user B's dataset DS B ,like Figure 3 As shown in the diagram. The result of the called procedure P is returned to user B's client device, as shown in the diagram. Figure 3 As shown in the image.

[0052] In a further possible embodiment, the invoked process P can be executed both based on the generated dataset pool that provides the first result and based on the current dataset DS uploaded by the user's client device, to provide only further results. The difference between these two results can form a benchmark indicating an increase in efficiency in processing the uploaded dataset DS via the invoked process, an increase in efficiency caused by data aggregation. In this embodiment, the user can recognize the efficiency increase caused by data aggregation, and will be more likely to accept data aggregation the next time a recommendation message REC is received from cloud platform 1. In a possible embodiment, the efficiency increase can be calculated by cloud platform 1 and provided to the user's client device for display to the user via the client device's user interface. Most analytical processes are designed to benefit from as much data as possible, and these processes can be performed more efficiently if many datasets from different users are aggregated in a data pool. Data aggregation is particularly beneficial when performing the training process of machine learning data models, especially artificial neural networks (ANNs). In a possible embodiment, cloud platform 1 can monitor users building machine learning models on the federated platform and inform them of the benefits of aggregating data. To enhance security, the provided data can be encrypted before being shared with other users in the data pool. In a preferred embodiment, the provided user data undergoes homomorphic encryption before being aggregated into the data pool. This computer implementation method allows for the automatic sharing of user data on cloud platform 1. In possible embodiments, the aggregated data is used to train a data model, particularly an artificial neural network (ANN). The data model is trained based on data that, in possible embodiments, can be defined by a recipe, which may take the form of a Docker container, shell script, KNIME workflow, etc.

[0053] Analysis services or processes P can be provided by cloud platform 1 to train and / or build data models by executing these recipes on a specified dataset DS. In a possible embodiment, the computer implementation method according to the invention can automatically retrain the user's data model on the combined aggregated dataset DS. Cloud platform 1 provides a feedback mechanism that informs the user of the expected or realized benefits of pooling their data. A data evaluation engine implemented in the scoring calculation unit 2A can measure the level of similarity between datasets DS and determine whether it is meaningful to combine them. In a possible embodiment, each dataset DS uploaded by a user's client device of cloud platform 1 can compute and combine two numerical vectors. The first vector V contains the most common statistical features, such as the mean or standard deviation of a given dataset DS. The second vector can contain features highly relevant to the corresponding process P available on cloud platform 1. Whenever a new process P is introduced to cloud platform 1, the second vector can be enhanced with additional features.

[0054] Each pair of datasets DS uploaded by different users of cloud platform 1 can have a similarity score SS calculated based on its representation vector. Many different similarity measures exist that can be used by the score calculation unit 2A for this purpose, such as cosine similarity. Whenever the pairwise similarity between two datasets DS exceeds a configurable threshold, the score calculation unit 2A can represent it as a matching dataset DS.

[0055] Whenever cloud platform 1 detects a match, there is reason to believe that the datasets DS that these users are working on are similar. Cloud platform 1 can then compare the procedures P that these users typically invoke on their data. If a match is detected, it is clear that both users can benefit from sharing data between them. Cloud platform 1 can then issue recommendation RECs to both users to share their data, thereby improving the quality of the results of the executed procedures. For example, if the executed procedure P is a training procedure, the quality of the machine learning-based data model will increase.

[0056] Optionally, before releasing the recommended REC, cloud platform 1 can retrain some of the user's previously created data models on the combined dataset DS that aggregates the data to quantify the improvements in the quality of these data models.

[0057] If users accept the recommended REC for sharing the dataset DS, then cloud platform 1 can perform the same action on the combined dataset DS whenever one of the users invokes the corresponding procedure P. The result of the executed procedure P can become available to both users involved. To ensure user data privacy, homomorphic encryption is applied to the dataset DS before sharing the data. Alternatively, the execution of procedures such as training a data model can be performed in a secure environment of cloud platform 1 where the data is not available to the users. In a possible embodiment, cloud platform 1 can automatically identify users who will benefit from sharing data or information contained in their dataset DS. Cloud platform 1 provides incentives for users to upload and share their data. For example, shared data for training a data model results in a better data model for users who require less labeled data. Since collecting labeled data is expensive and time-consuming, cloud platform 1 significantly increases efficiency in training data models.

Claims

1. A computer implementation method for efficiently processing aggregated data shared by users of a cloud platform, the method comprising the following steps: - At least one dataset is uploaded to the cloud platform by the user's client device; - Calculate a similarity score, which indicates the degree of similarity between the currently uploaded dataset and datasets previously uploaded by other users' client devices; as well as - Perform user-selected analytics processes on the cloud platform based on aggregated data, which includes the user's current dataset and datasets previously uploaded from other users' client devices and stored in the cloud platform's database. The previously uploaded datasets have a calculated similarity score relative to the user's currently uploaded dataset that exceeds a configurable similarity score threshold. The dataset includes sensor data generated by sensors on machines in the respective users' industrial systems. Client devices of users who have already uploaded the current dataset to the cloud platform receive recommendation messages from the cloud platform to aggregate datasets from other users on the cloud platform that match the current dataset. If the cloud platform receives an acceptance message from the user's client device to aggregate datasets, it automatically aggregates the matching datasets to generate a dataset pool. The cloud platform performs a user-selected analysis process based on both a pool of generated datasets and a separate current dataset to calculate a benchmark based on the difference between the results of the execution based on the pool of generated datasets and the results based on the separate current dataset. This benchmark indicates an increase in efficiency in processing the uploaded dataset through the selected analysis process, the increase in efficiency being caused by data aggregation.

2. The method according to claim 1, wherein the analysis process performed by the cloud platform includes: The training process used to train data models. The testing process used to test the trained data model, and / or The reasoning process used to execute a trained and tested data model.

3. The method of claim 2, wherein the data model used includes an artificial neural network (ANN).

4. The method according to claim 1 or 2, wherein for each uploaded dataset, a representation vector is computed, the representation vector comprising vector elements representing the statistical properties of the uploaded dataset.

5. The method of claim 4, wherein a similarity score indicating the degree of similarity between the uploaded datasets is calculated based on the representation vector of the uploaded dataset.

6. The method of claim 1 or 2, wherein the calculated similarity score includes a cosine similarity score.

7. The method of claim 1 or 2, wherein if the similarity score calculated for the currently uploaded dataset relative to previously uploaded datasets of other users exceeds a configurable similarity score threshold, the corresponding previously uploaded and stored dataset is marked as a matching dataset for the dataset currently uploaded by the user's client device.

8. The method of claim 7, wherein the aggregated data includes datasets uploaded from client devices of different users and tagged as matching datasets.

9. The method of claim 1 or 2, wherein a similarity score is triggered to calculate datasets previously uploaded by other users and stored in a database on the cloud platform in response to the upload of a new current dataset from the client device of the corresponding user.

10. The method of claim 1 or 2, wherein the matching datasets undergo homomorphic encryption before they are aggregated to generate a dataset pool.

11. The method of claim 1 or 2, wherein the calculated benchmark is sent by the cloud platform to the user's client device.

12. The method of claim 1 or 2, wherein the dataset comprises labeled data and / or unlabeled data.

13. A cloud platform for efficiently processing aggregated data shared by users of a cloud platform, the cloud platform comprising: - A database that is adapted to store datasets uploaded to the cloud platform by users' client devices; - A scoring unit configured to calculate similarity scores, each similarity score indicating the degree of similarity between the currently uploaded dataset and previously uploaded datasets stored in the database; as well as - A processor adapted to perform a user-selected analysis process based on aggregated data, which includes the current dataset uploaded by the corresponding user and previously uploaded datasets from other users, the previously uploaded datasets having calculated similarity scores that exceed a configurable similarity threshold relative to the current dataset uploaded by the corresponding user. The dataset includes sensor data generated by sensors on machines in the respective users' industrial systems. Client devices of users who have already uploaded the current dataset to the cloud platform receive recommendation messages from the cloud platform to aggregate datasets from other users on the cloud platform that match the current dataset. If the cloud platform receives an acceptance message from the user's client device to aggregate the dataset, it automatically aggregates the matching datasets to generate a dataset pool. The cloud platform performs a user-selected analysis process based on both a pool of generated datasets and a separate current dataset to calculate a benchmark based on the difference between the results of the execution based on the pool of generated datasets and the results based on the separate current dataset. This benchmark indicates an increase in efficiency in processing the uploaded dataset through the selected analysis process, the increase in efficiency being caused by data aggregation.

Images