Method and system for operating a technical apparatus

Abstract

The invention relates to a method for operating a technical apparatus by means of a system comprising the technical apparatus, an operating device for operating the technical apparatus, and a computer system designed to carry out a first autoencoder (4) and a second autoencoder (8), each on the basis of a neural network, and to carry out the following steps: a) training the first autoencoder (4) and the second autoencoder (8) with respective input data, wherein provided training data (6) is used as input data for the first autoencoder (8) and the output data (10) of the first autoencoder (4) is used as input data of the second autoencoder (8), b) determining reconstruction error between the training data (6) and output data (12) of the second autoencoder (8), and c) continuing with step a) so long as a measure of the size of the reconstruction error exceeds a predefined value, otherwise continuing with step d), d) providing operating data as input data for the first autoencoder (8), and transmitting the output data (10) of the first autoencoder (4) as input data of the second autoencoder (8), and operating the technical apparatus by means of the operating device on the basis of the output data (12) of the second autoencoder (8).

Description

[0001] Method and system for operating a technical device

[0002] The invention relates to a method and a system for operating a technical device, comprising the technical device, an operating device for operating the technical device, and a computer system which is configured to have a first autoencoder and a second autoencoder, each based on a neural network.

[0003] The invention further relates to a computer program product and a machine-readable storage medium.

[0004] Machine learning algorithms, abbreviated to ML algorithms, are generally based on the use of statistical methods to train a data processing system so that it can perform a specific task without it having been explicitly programmed to do so. The goal of machine learning is to construct algorithms that can learn from data and make predictions. ML applications represent a subgroup of AI ("artificial intelligence") applications. In general, the abbreviations "AI" stand for "artificial intelligence", "AI" for the corresponding English term "artificial intelligence" and "ML" for "machine learning" or the corresponding English term "machine learning".

[0005] Before using or applying such machine learning algorithms, they are trained based on the training data characterizing the respective application case, whereby, for example, weights within the machine learning algorithm are automatically adjusted in such a way that the machine learning algorithm is increasingly able to identify relationships between features and predictions or

[0006] input data and corresponding output data.

[0007] Recently, cloud-based structures have been increasingly used for data storage and analysis. Typically, in cloud-based ML and AI applications (machine learning and artificial intelligence applications), all raw data is transferred to the cloud and stored in an optimized database (e.g., a file or object store). AI models are then trained on this data by retrieving the raw data and running the ML algorithms on optimized computing hardware. If a data leak occurs (e.g., due to the inadvertent use of incorrect security permissions for a bucket), sensitive data such as images of defective workpieces can be exposed to the internet. This can lead to enormous reputational damage for the company whose data has been leaked.There is also further potential for misuse through data that can be downloaded or accessed in any other way without authorization.

[0008] To avoid the possibility of data leaks described above, it is possible, for example, to forego the use of a cloud-based infrastructure for sensitive data. However, this also means foregoing the benefits of the elastic scaling capabilities of a cloud-based computer cluster.

[0009] Another option is to encrypt the storage or raw data to ensure the protection of sensitive data. Aside from other known problems associated with such a method, the cloud provider typically manages the encryption keys. This means that security-relevant processes are forcibly outsourced, which, at least theoretically, could lead to further risk factors regarding the possibility of unauthorized access to sensitive data.

[0010] The object of the present invention is to provide a way to secure the use of a cloud-based infrastructure when training ML and AI applications against unauthorized access to the data used.

[0011] The task is also to further increase the accuracy of a model based on artificial intelligence in a simple way, as well as to advantageously improve the data protection of the data used.

[0012] According to the invention, a method for training an AI system for anonymizing data is provided, which makes it possible to make data exchange with a cloud-based infrastructure more secure with regard to unauthorized access to transferred data by third parties.

[0013] A computer program and a machine-readable storage medium are also provided to solve the problem.

[0014] The object of the invention is thus achieved by a method for operating a technical device by a system comprising the technical device, an operating device for operating the technical device, and a computer system which is configured to have a first autoencoder and a second autoencoder, each based on a neural network, and to carry out the following steps: a) training the first autoencoder and the second autoencoder with respective input data, wherein provided training data are used as input data for the first autoencoder, and the output data of the first autoencoder are used as input data of the second autoencoder, b) determining a reconstruction error between the training data and output data of the second autoencoder, and c) continuing with step a) as long as a measure of the size of the reconstruction error exceeds a predetermined value, otherwise continuing with step d),d) Providing operating data as input data for the first autoencoder and applying the first model, and transmitting the output data of the first autoencoder as input data of the second autoencoder and applying the second model, and transmitting the output data of the second autoencoder to the operating device, and operating the technical device by the operating device using the output data of the second autoencoder.

[0015] It is clear that the method according to the invention is provided with the appropriate training data and / or output models for the autoencoders.

[0016] Furthermore, it is clear that the components of the system, namely the technical device, the operating device for operating the technical device, and the computer system are designed to communicate with each other accordingly and are connected accordingly.

[0017] It follows from the method steps that during training the training data are compared with the output of the second autoencoder, whereby the comparison can be carried out on the input side of the first autoencoder or on the output side of the second autoencoder.

[0018] It may be necessary - temporarily - to transmit the training data or the output data of the second autoencoder, preferably as encrypted data.

[0019] During inference, i.e. when the operational data is used as input data for the first autoencoder, this data transmission can be deactivated, since both autoencoders are set up to each other and perform an anonymization or deanonymization of the operational data, and only transmit anonymous data between system units, such as different clients.

[0020] The method according to the invention can be carried out, for example, as follows: An original image or another set of raw data is made available to the first autoencoder as a training data set. The autoencoder can in particular be a type of artificial neural network that can be used to learn efficient encodings of unlabeled data (unsupervised learning). An autoencoder usually learns two functions: an encoding function that transforms the input data, and a decoding function that recreates the input data based on the encoded representation. The autoencoder learns an efficient representation (encoding) for a data set, typically for dimensionality reduction. The representation of the dimensionally reduced data set takes place in a so-called latent space, which can also be referred to as a "bottleneck". The AI ​​system can be an ML system.

[0021] In a further development of the invention, it is provided that a dist() function is used to determine the reconstruction error.

[0022] The first autoencoder, in turn, contains the first encoder, the first latent space, and the second decoder. The training dataset can be sent to the first decoder through the first encoder and the first latent space. In an early stage of the learning process, the reconstruction error (the difference between the original image and the reconstructed image, or the difference between the original dataset and the reconstructed dataset) will be high because the first autoencoder has not yet learned a sufficient representation of the dataset or image.

[0023] In a further development of the invention, it is provided that the system is a client-server system with a server and a first client, and the computer system has a first computing device which is comprised by the first client, and the first computing device is configured to execute the first autoencoder.

[0024] This ensures that only the first client has knowledge of raw data, i.e. training data and / or operating parameters, and not the server, thus ensuring a high level of data protection and privacy.

[0025] In a further development of the invention, it is provided that the client-server system comprises a second client, and a second computing device of the computer system is comprised by the second client, and the second computing device is configured to execute the second autoencoder.

[0026] This ensures that the second client uses anonymous data, i.e. only a model structure and model weights of the neural network in the form of the autoencoder, when applying the second autoencoder, whereby the second client has no knowledge of raw data, i.e. training data and / or operating parameters, thus ensuring an even higher level of data protection and privacy.

[0027] The reconstructed image or the reconstructed data set can then be sent through the second autoencoder, which, like the first autoencoder, contains a second encoder, a second latent space and a second decoder, and compared with the original input image or with the original data set that was input to the first autoencoder. The reconstruction error can be measured with the function dist(), which is known, for example, from the R programming language. As soon as the distance is sufficiently small, training is terminated. The termination criterion for training the first autoencoder and the second autoencoder can therefore be based on the reconstruction error of the second autoencoder or can be formed from this reconstruction error. The first autoencoder and the second autoencoder can be trained jointly or building on each other.The first autoencoder is inherently incapable of creating a perfect representation of every input image in the dataset. Instead, it will generate a noisy dataset that must be decoded by the second autoencoder. The second autoencoder therefore learns to reconstruct the original dataset from noisy images.

[0028] The object according to the invention is also achieved by a system for operating a technical device, comprising the technical device and at least one computer system, wherein the system is set up to carry out the method according to the invention.

[0029] In a further development of the invention, it is provided that the system is a client-server system with a server and a first client, and the computer system has a first computing device which is comprised by the first client, and the first computing device is configured to execute the first autoencoder.

[0030] In a further development of the invention, it is provided that the client-server system comprises a second client, and a second computing device of the computer system is comprised by the second client, and the second computing device is configured to execute the second autoencoder.

[0031] The provided method has the advantage that the anonymization of data or data sets can be carried out in an energy- and computing-power-efficient manner. Furthermore, the approach of the provided method can be embedded in an ML pipeline. This eliminates the need for a separate application, for example, for encryption or another type of anonymization. Furthermore, bandwidth and storage space can be saved because data to be transmitted to the cloud-based infrastructure for further processing can be transmitted at a lower resolution. This applies both to image data, for which the term "resolution" can be taken literally, and to other data that typically requires less storage after anonymization.

[0032] Anonymizing data can be understood, in particular, as a process that modifies data in such a way that its usability for third parties is reduced. Deanonymization is the reverse process, in which the original data is reconstructed as completely as possible. The original data and the initially anonymized and then deanonymized data are generally not identical, as a loss of information content may occur.

[0033] The method according to the invention generates data or a model that can be used, for example, to operate a technical device. During operation, the obtained data, which lie within a predetermined, permissible reconstruction error, can therefore be used to control a device with a control device or to read data from a device and evaluate it accordingly, for example, for reconstructing measurement data, as previously described using image data.

[0034] It is clear that the method according to the invention is provided with operating data for the operation of the technical device or is recorded by a corresponding sensor means.

[0035] In other words, in the method according to the invention the first autoencoder of the AI ​​system is used with the help of a first computer system, such as a control device, to operate a technical device connected to the AI ​​system. Furthermore the method according to the invention enables the owner of the data to retain full control over the data set, since the data set is useless without the deanonymization model. If the data consists of image data or other measurement data, for example, the data set in the cloud will consist of blurred images or insufficiently sampled data points. As long as the deanonymization model is not intercepted in parallel with the confidential data or otherwise reaches unauthorized users, the anonymized data cannot be used meaningfully. The content of the data therefore remains secure and cannot be reconstructed by third parties.

[0036] In an advantageous embodiment, the first autoencoder and the second autoencoder each comprise an encoder module and a decoder module. Thus, well-known principles and software modules can be used to implement the method.

[0037] It is advantageous to use a dist() function to determine the reconstruction error. The dist() function can, for example, be the dist() function of the R programming language. Here, too, it is advisable to use already known and proven building blocks in order to reduce the overall effort required to implement the method according to the invention.

[0038] According to one embodiment, the method for training an AI system for anonymizing data is carried out on an edge device. An edge device can in particular be understood to be a device that represents an access point to the core networks of companies or service providers. Examples of this are routers, routing switches, integrated access devices (IADs), multiplexers and a large number of MAN (Metropolitan Area Network) and WAN (Wide Area Network) access devices. Edge devices can also establish connections to operator and service provider networks. The edge device can be directly and thus securely connected to the data source that supplies the raw data. This data source can, for example, be a camera that supplies images to monitor the quality of a manufactured product. Other data sources are of course also possible.The raw data can therefore be used directly after creation to train the AI ​​system for anonymizing data.

[0039] The method can expediently be carried out on a computer system using an artificial intelligence runtime app or an AI runtime app that can execute machine learning packages or ML packages. This computer system can be the edge device described above. This often allows the use of an existing hardware and software environment, saving development, setup, maintenance, and acquisition costs.

[0040] An advantageous extension of the method described above consists in a method for training an AI application, comprising the steps of a. anonymizing raw data using a first computer system with a first autoencoder which was trained using the method described above, b. transferring the anonymized raw data via a network to a second computer system, c. deanonymizing the anonymized raw data using the second computer system with a second autoencoder which was trained using a method described above, and d. training the AI ​​application with the deanonymized raw data.

[0041] According to an advantageous embodiment, the second computer system is part of a cloud architecture. This makes it possible to utilize the advantages of the elastic scaling capabilities of a cloud-based computer cluster while simultaneously maintaining high data security.

[0042] The first computing system can be an edge device. It can be the same edge device that was used to train the AI ​​system to anonymize data—in other words, to train the first autoencoder and the second autoencoder.

[0043] It is possible for the second encoder to be stored on the second computer system. In other words, the second encoder can be stored, in particular, in a cloud architecture. Deanonymization can then be performed in the immediate vicinity of the data analysis device or the cloud-based ML or AI application that is intended to use the data. Accordingly, the risk of data leaks or other unauthorized access to the raw data in usable form is further minimized.

[0044] Using the described method, it is possible to train an AI model in a cloud environment, whereby at no time is usable raw data processed or stored in the cloud outside of the AI ​​runtime. The AI ​​runtime can be a proprietary Docker container. Without the ML package for deanonymization provided via a secure channel, for example via an ML package store, it is not possible to use the data meaningfully. The ML package for deanonymization can contain the second autoencoder, which can be used to restore the usability of the data.

[0045] The method according to the invention generates a trained AI model, which can be used, for example, to operate a technical device. During operation of the device, the obtained data, which lie within a predetermined, permissible reconstruction error, are used to control a device with a control device or to read data from a device and evaluate it accordingly, for example, for reconstructing measurement data, as previously described using image data.

[0046] The method according to the invention can be provided with operating data for the operation of the technical device or can be recorded by a corresponding sensor means.

[0047] In other words, in the method according to the invention, the trained artificial intelligence application is used with the aid of a control device to operate a technical device connected to the AI ​​system.

[0048] The simplest application is that the two autoencoders are only used to provide data transmission of anonymous data between the two autoencoders.

[0049] In addition, a Kl function can be taken into account in the two autoencoders during the training phase, for example for anomaly detection, whereby only "error-free" data records are taken into account in the training data and "error-prone" data records are excluded from the training data.

[0050] For example, it can therefore be provided that image data of a product manufactured by a production plant is captured by a camera and provided to the first autoencoder as input data, and the image data is fed to an AI-based evaluation with the help of a corresponding AI model for anomaly detection in the manufactured product in order to detect a possible production error or inadmissible product accuracy, in order to then adjust the production plant accordingly and thus achieve an improved production result, whereby an improved operation of the production plant can be achieved.

[0051] However, it can also be provided that, for example, operating data from a pump is recorded by a corresponding sensor and provided to the first autoencoder as input data. The operating data is then analyzed by an AI-based evaluation using a corresponding AI model for anomaly detection during pump operation in order to identify a possible operating error or impermissible operating accuracy, in order to then control the pump accordingly and thus achieve an improved operating result. Anomaly detection can preferably be used in the training phase in order to improve or make anomaly detection during operation more precise.

[0052] The operation of the technical device can, for example, be controlled by appropriate control information from the operating equipment. Alternatively, a simple alarm or signaling information can be output regarding the anomaly detection. Subsequently, for example, a connected sorting device can reject a manufactured product that has been detected as incorrect.

[0053] The above-described properties, features, and advantages of this invention, as well as the manner in which they are achieved, will become clearer and more readily understood in connection with the following description of the embodiments, which are explained in more detail in connection with the drawings.

[0054] Figure 1 is a flow chart illustrating a first embodiment of a first part of the method according to the invention,

[0055] Figure 2 is a flow chart illustrating a second embodiment of a further part of the method according to the invention,

[0056] Figure 3 shows a schematic overview of a hardware and network architecture suitable for carrying out a method according to the invention,

[0057] Figure 4 is a schematic representation of a data flow according to an embodiment of a method according to the invention for training a Kl system for anonymizing data, and

[0058] Figure 5 shows a schematic representation of a data flow according to a further embodiment of a method according to the invention. Figure 1 shows a flowchart illustrating a first embodiment of a first part of the method according to the invention. The method can accordingly be a method for training an AI system for anonymizing data, and can be used for operating a technical device connected to the AI ​​system with the generated data or the trained AI model.

[0059] In a step S100, a first autoencoder and a second autoencoder of a Kl system for anonymizing data are trained with training data, wherein the training data are used as input data for the first autoencoder, and wherein the output data of the first autoencoder are used as input data for the second autoencoder.

[0060] In step S102, a reconstruction error between the training data and the output data output by the second autoencoder is determined. The determined reconstruction error can then be compared with a predetermined reference value, or a difference can be calculated between the predetermined reference value and the reconstruction error.

[0061] In a step S104, training is terminated if a measure of the magnitude of the reconstruction error falls below a predetermined value. For this purpose, the result of the previously performed comparison or the previously determined difference can be evaluated. If it is determined that the reconstruction error is still too large, in other words, if the reconstruction of the training data provided by the second autoencoder still deviates too greatly from the original training data, training is continued. For this purpose, the method can be returned to step S100.

[0062] Figure 2 shows a flowchart illustrating a second embodiment of a further part of the method according to the invention. The method can be a method for training an ML application.

[0063] In a step S200, raw data are anonymized using a first computer system with a first autoencoder which has been trained using a method according to the invention, for example the method described in connection with Figure 1.

[0064] In a step S202, the anonymized raw data are transferred anonymously, i.e. without the operating data itself, via a network to a second computer system, whereby this is done exclusively by the output data of the second autoencoder.

[0065] The transmitted data is thus a "fuzzy" representation of the operating data and is suitable for data transmission between individual clients, i.e. edge devices, or to the server of the client-server system.

[0066] A loss of this "fuzzy data" is usually unproblematic, since no special knowledge is contained, or is only available after deanonymization by the second autoencoder.

[0067] According to the invention, it is therefore only possible to retrieve the complete information content of the operating data if the model weights and the model structure of the second autoencoder are known. The model weights and the model structures of the first and second autoencoders are generated during training and therefore do not need to be transmitted separately, which, for example, eliminates the need for encrypted transmission.

[0068] In a step S204, the anonymized raw data are deanonymized using a second computer system with a second autoencoder, which was trained with a method according to the invention, for example with the method described in connection with Figure 1, wherein the "deanonymization", i.e. the reconstruction of the input data of the first autoencoder, from the output data of the first autoencoder is carried out with the aid of the second autoencoder.

[0069] In step S206, the AI ​​model is then trained with the now deanonymized raw data. The AI ​​model can thus be trained with high-resolution or high-quality data without exposing this data to security risks during network transfer.

[0070] Figure 3 shows a schematic overview of a hardware and network architecture in the form of a K1 system 2, which is suitable for implementing a method according to the invention. The inventive concept is illustrated using image data, but can also be applied to other data types (e.g., time series or tabular data).

[0071] The data flow can be described as follows: A camera acts as a data source 26 and collects images from a production line (not shown) (e.g. for visual quality applications). The edge device 28, which can also be referred to as an edge device, represents a first computer system 30 and processes the data locally. It hosts an AI runtime app 32 that can execute ML packages 34. An ML package 34 is understood in particular to be an artifact that contains the code for machine learning structured as an ML pipeline. The ML pipeline can contain modules for data preprocessing, training, inference and post-processing of the data. Such an ML package 34 can be the first encoder, which can then be used to anonymize the data.

[0072] Instead of the camera 26 as a technical device, in an alternative embodiment, which is not shown, a robot can also be used as a Kl application for a process control with control data as operating data with the aid of an autoencoder.

[0073] The edge device 28 then sends the anonymized data to the cloud 42 or to the object storage or data storage 36 located there. The AI ​​runtime app 44, which can execute ML packages 46, can be executed on a second computer system 40, such as an ML training service 48 in the cloud, for example, on a service offered by a third-party provider. The ML package for data de-anonymization, which can be generated by the second autoencoder, can be provided from a local repository or from a local data storage or program storage 38 directly in the AI ​​runtime container.

[0074] Figure 4 shows a schematic representation of a data flow according to an embodiment of a method according to the invention for training a KI system for anonymizing data. First, training data 6 are input to a first autoencoder 4. In simple terms, the first autoencoder 4 consists of a first encoder 14, a first latent space 16 and a first decoder 18. The training data are visualized as an image with a relatively sharp representation. The first autoencoder 4 outputs first output data 10. The quality of the output data 10 is of lower resolution or noisy, which is indicated by the poorer quality representation of the first output data 10 compared to the original training data 6. The first output data 10 are then input to the second autoencoder 8.The second autoencoder 8 also has a second encoder 20, a second latent space 22 and a second decoder 24.

[0075] The second autoencoder 8 outputs second output data 12 . The quality of the second output data 12 varies depending on how far the training has progressed. In the case shown, the resolution of the second output data 12 is similarly low to the resolution of the first output data 10 . The second output data 12 is compared with the training data 6, and a reconstruction error is determined. A dist() function can be used for this purpose. Depending on the determined reconstruction error, the training can be continued or terminated.

[0076] Figure 5 shows a schematic representation of a data flow according to an embodiment of a method according to the invention for training an ML application, wherein the method is based on the use of the previously trained first autoencoder 4 and the second autoencoder 8 trained jointly with the first autoencoder 4. Raw data 50 is now input to the first autoencoder 4 for processing. The first autoencoder 4 outputs anonymized raw data 54, which is input to the second autoencoder 8 for further processing. The second autoencoder 8 deanonymizes the input data and outputs the deanonymized raw data 58.

[0077] The first autoencoder 4 can be stored on or executed by a first computer system 52, and the second autoencoder 8 can be stored on or executed by a second computer system 56. The first computer system 52 is preferably arranged in the immediate vicinity of the generation location of the raw data 50. The second computer system 56 can be connected to the first computer system 52 via a network connection (not shown) and / or can be integrated into a cloud architecture, for example.

[0078] Although the invention has been illustrated and described in detail by means of preferred embodiments, the invention is not limited by the disclosed examples and other variations can be derived therefrom by those skilled in the art without departing from the scope of the invention.

[0079] List of reference symbols

[0080] 2 Kl - System, System, Client-Server-System

[0081] 4 first autoencoder

[0082] 6 Training data

[0083] 8 second autoencoder

[0084] 10 first output data

[0085] 12 second output data

[0086] 14 first encoder module

[0087] 16 first Latent Space

[0088] 18 first decoder module

[0089] 20 second encoder module

[0090] 22 second latent space

[0091] 24 second decoder module

[0092] 26 Data source

[0093] 28 Edge Device , Client

[0094] 30 first computer system, first computing device

[0095] 32 AI runtime app

[0096] 34 ml package

[0097] 36 data storage

[0098] 38 program memories

[0099] 40 second computer system

[0100] 42 Cloud, Server System

[0101] 44 AI runtime app

[0102] 46 ml package

[0103] 48 ML training service

[0104] 50 raw data

[0105] 52 first calculation system

[0106] 54 anonymized raw data

[0107] 56 second calculation system

[0108] 58 deanonymized raw data

Claims

Patent claims 1. A method for operating a technical device by a system comprising the technical device, an operating device for operating the technical device, and a computer system which is configured to have a first autoencoder (4) and a second autoencoder (8), each based on a neural network, and to carry out the following steps: a) training the first autoencoder (4) and the second autoencoder (8) with respective input data, wherein provided training data (6) are used as input data for the first autoencoder (4), and the output data (10) of the first autoencoder (4) are used as input data of the second autoencoder (8), b) determining a reconstruction error between the training data (6) and output data (12) of the second autoencoder (8), and c) continuing with step a) as long as a measure of the size of the reconstruction error exceeds a predetermined value, otherwise continuing with step d),d) Providing operating data as input data for the first autoencoder (8) and applying the first model, and transmitting the output data (10) of the first autoencoder (4) as input data of the second autoencoder (8) and applying the second model, and transmitting the output data (12) of the second autoencoder (8) to the operating device, and operating the technical device by the operating device using the output data (12) of the second autoencoder (8).

2. Method according to the preceding claim, characterized in that a dist () function is used to determine the reconstruction error.

3. The method according to any one of the preceding claims, wherein the system is a client-server system having a server and a first client, and the computer system comprises a first computing device (30) which is comprised by the first client (28), and the first computing device (30) is configured to execute the first autoencoder (4).

4. The method according to the preceding claim, wherein the client-server system comprises a second client, and a second computing device (40) of the computer system is comprised by the second client (28), and the second computing device (40) is configured to execute the second autoencoder (8).

5. System for operating a technical device, comprising the technical device and at least one computer system, wherein the system is configured to carry out the method according to one of the preceding steps.

6. System according to the preceding claim, wherein the system is a client-server system with a server and a first client, and the computer system has a first computing device (30) which is comprised by the first client (28), and the first computing device (30) is configured to execute the first autoencoder (4).

7. System according to the preceding claim, wherein the client-server system comprises a second client, and a second computing device (40) of the computer system is comprised by the second client (28), and the second computing device (40) is configured to execute the second autoencoder (8).

8. A computer program product comprising instructions that are Execution of the contained program by a computer causes the computer to carry out a method according to one of claims 1 to 4.

9. A machine-readable storage medium on which the computer program according to claim 8 is stored.

Images