Computer-implemented method for grounding a machine learning model

Abstract

A computer-implemented method (100) for grounding a machine learning model comprises: obtaining (101) a source dataset, wherein the source dataset comprises a plurality of images; generating (102) a description of each image in the source dataset; generating (103) a grounding dataset by selecting descriptions of images in the source dataset to the grounding dataset based at least on a classification of each image in the source dataset and a number of characters in each image in the source dataset; and using (104) the grounding dataset for grounding a machine learning model.

Description

COMPUTER- IMPLEMENTED METHOD FOR GROUNDING A MACHINE LEARNING MODELTECHNICAL FIELD

[0001] The present disclosure relates to machine learning, and more particularly to a computer-implemented method for grounding a machine learning model , a computing device , and a computer program product .BACKGROUND

[0002] Machine learning models , such as large langue models , can be utili zed in, for example , answering questions presented by users . For example , a company can utili ze a large machine learning model to answers questions about their services and to provide technical support to users . In such applications , it can be benefi cial to configure the machine learning model to provide the answers according to a custom dataset , such as technical documentation, instead of generic training information . This may be referred to as grounding .SUMMARY

[0003] This summary is provided to introduce a selection of concepts in a s implif ied form that are further described below in the detailed description . This summary is not intended to identify key features or essential features of the claimed subj ect matter, nor is itintended to be used to limit the scope of the claimed subj ect matter .

[0004] It is an obj ective to provide a computer-implemented method for grounding a machine learning model , a computing device , and a computer program product . The foregoing and other obj ectives are achieved by the features of the independent claims . Further implementation forms are apparent from the dependent claims , the description and the figures .

[0005] According to a first aspect, a computer-implemented method for grounding a machine learning model , the method comprising : obtaining a source dataset , wherein the source dataset comprises a plurality of images ; generating a description of each image in the source dataset ; generating a grounding dataset by selecting descriptions of images in the source dataset to the grounding dataset based at least on a classification of each image in the source dataset and a number of characters in each image in the source dataset ; and using the grounding dataset for grounding a machine learning model .

[0006] In an implementation form of the first aspect , the generating the description of each image in the source dataset comprises generating the description of each image in the source dataset using an image description machine learning model .

[0007] In another implementation form of the first aspect , the method further comprises classifying eachimage in the source dataset using an image classi fication machine learning model , thus obtaining the classification of each image in the source dataset .

[0008] In another implementation form of the first aspect , the method further comprises determining the number of characters in each image in the source dataset using a character recognition machine learning model .

[0009] In another implementation form of the first aspect , the class ification of each image in the source dataset classifies the image to at least one image class from a plurality of image classes , wherein the plurality of image classes comprises a non-inf ormative image class and at least one informative image class .

[0010] In another implementation form of the first aspect , the selecting descriptions of images in the source dataset to the grounding dataset based at least on the classification of each image in the source dataset and the number of characters in each image in the source dataset comprises selecting descriptions of images in the source dataset to the grounding dataset based on whether an image comprises more characters than a threshold number of characters .

[0011] In another implementation form of the first aspect , the selecting descriptions of images in the source dataset to the grounding dataset based at least on the classification of each image in the source dataset and the number of characters in each image in the source dataset comprises selecting descriptions of images in the source dataset to the grounding datasetbased on whether an image belongs to a predetermined image class .

[0012] In another implementation form of the first aspect , the selecting descriptions of images in the source dataset to the grounding dataset based at least on a classification of each image in the source dataset and a number of characters in each image in the source dataset comprises , in response to an image comprising more characters than a threshold number of characters and the image belonging to a predetermined image class , selecting the description of the image to the grounding dataset .

[0013] In another implementation form of the first aspect , the using the grounding dataset for grounding the machine learning model comprises providing a prompt to the machine learning model , wherein the prompt comprises a request and instructions to the machine learning model to respond to the request using the grounding dataset .

[0014] In another implementation form of the first aspect , at least one image in the source dataset comprises a plurality of sub images ; the generating the description of each image in the source dataset comprises generating a description of each sub image in the plurality of sub images ; and the generating the grounding dataset by selecting descriptions of images in the source dataset to the grounding dataset comprises selecting descriptions of sub image in the plurality of sub images to the grounding dataset based at least on aclassification of each sub image in the plurality of sub images and a number of characters in each sub image in the plurality of sub images .

[0015] According to a second aspect , a computing device compri ses at least one processor and at least one memory including computer program code , the at least one memory and the computer program code being configured to , with the at least one proces sor, cause the computing device to perform the method according to the first aspect .

[0016] According to a third aspect , a computer program product comprises program code configured to perform the method according to the first aspect when the computer program product is executed on a computer .

[0017] Many of the attendant features wil l be more readily appreciated as they become better understood by reference to the following detailed description considered in connection with the accompanying drawings .DESCRIPTION OF THE DRAWINGS

[0018] In the following, example embodiments are described in more detail with reference to the attached figures and drawings , in which :

[0019] Fig . 1 illustrates a flow chart representation of a method according to an embodiment ;

[0020] Fig . 2 illustrates a flow chart representation of a method according to another embodiment ;

[0021] Fig . 3 illustrates a flow chart representation of a method according to another embodiment ;

[0022] Fig . 4 illustrates a schematic representation of image classes according to an embodiment ; and

[0023] Fig . 5 illustrates a schematic representation of a computing device according to an embodiment .

[0024] In the following, like reference numerals are used to des ignate li ke parts in the accompanying drawings .DETAILED DESCRIPTION

[0025] In the following description, reference is made to the accompanying drawings , which form part of the disclosure , and in which are shown, by way of illustration, specific aspects in which the present disclosure may be placed . It is understood that other aspects may be utilised, and structural or logical changes may be made without departing from the scope of the present disclosure . The following detailed description, therefore , is not to be taken in a limiting sense , as the scope of the present disclosure is defined by the appended claims .

[0026] For instance , it is understood that a disclosure in connection with a described method may also hold true for a corresponding device or system configured to perform the method and vice versa . For example , if a specific method step is described, a corresponding device may include a unit to perform the described method step, even if such unit is not explicitly described or il lustrated in the f igures . On the other hand, for example , if a specific apparatus is described based onfunctional units , a corresponding method may include a step performing the described functionality, even if such step is not explicitly described or illustrated in the figures . Further, it is understood that the features of the various example aspects described herein may be combined with each other, unless specifically noted otherwise .

[0027] Fig . 1 illustrates a flow chart representation of a method according to an embodiment .

[0028] According to an embodiment , a computer-implemented method 100 for grounding a machine learning (ML) model comprises obtaining 101 a source dataset, wherein the source dataset comprises a plurality of images .

[0029] The obtaining 101 the source dataset may comprise , for example , obtaining the source dataset from a database or any other storage / memory .

[0030] The source dataset may further comprise other data in addition to the plural ity of images . For example, the source dataset may comprise documents comprising text and images and / or documents comprising text without any images .

[0031] The source dataset may comprise , for example , technical documentation and / or other documentation . Such documentation can be pre-processed by the method 100 in order to be used for grounding a machine learning model .

[0032] The method 100 may further comprise generating 102 a description of each image in the source dataset .

[0033] The description of each image may be generated in various ways , such as those disclosed herein .

[0034] The method 100 may further comprise generating 103 a grounding dataset by selecting descriptions of images in the source dataset to the grounding dataset based at least on a classification of each image in the source dataset and a number of characters in each image in the source dataset .

[0035] The classification of each image and / or the number of characters in each image may be available , for example , in the source dataset . Alternatively, the classification of each image and / or the number of characters in each image may be available in some other manner or they may be generated using, for example , the procedures disclosed herein .

[0036] Herein, a character may refer to any written symbol , such as a letter or a numerical digit . A character may al so be referred to as a typographical mark, symbol , or similar .

[0037] The method 100 may further comprise using 104 the grounding dataset for grounding a machine learning model .

[0038] Grounding may refer to indicating to the machine learning model what information the machine learning model should use when responding to a request, such as a question, provided to the machine learning model . Grounding may be also referred to as retrieval-augmented generation (RAG) .

[0039] Herein a machine learning model may also be referred to as an artificial intelligence (Al ) model or similar .

[0040] According to an embodiment , the using 104 the grounding dataset for grounding the machine learning model comprises providing a prompt to the machine learning model , wherein the prompt comprises a request and instructions to the machine learning model to respond to the request using the grounding dataset .

[0041] For example , the prompt may comprise a system section / part and the system section / part may be used to instruct the machine learning model , such as a large language model (LLM) , to provide a response to the request , such as a question presented by a user, based on the grounding dataset .

[0042] According to an embodiment , the us ing 104 the grounding dataset for grounding the machine learning model comprises : obtaining a question from a user, providing a prompt to the machine learning model , wherein the prompt comprises the question from the user and instructions to the machine learning model to respond to the question using the grounding dataset .

[0043] For example , it can be advantageous that a chatbot uti li z ing a machine learning model , such as an LLM, answers questions from users based on a custom knowledge base comprising the grounding dataset instead of generic training information from, for example , the internet .

[0044] According to an embodiment, the machine learning model comprises a large language model .

[0045] For example , a chatbot utili zing an LLM to answer questions presented by users may utili ze the grounding dataset in order to provide accurate answers to the questions .

[0046] It should be appreciated that at least some of the operations of the method 100 may be performed in various orders . Further, in some embodiments , images in the source dataset may be processed one-by-one . For example , in some embodiments , an image can be obtained from the source dataset and the description can be generated for the image , it can be checked whether the descriptions of the image should be added to the grounding dataset based at least on a classification of the image and the number of characters in the image , and the description of the image can be added to the grounding dataset if the conditions are fulfilled . This can be repeated in an iterative manner for each image in the source dataset .

[0047] The method 100 can, for example , improve the utilisation of information comprised in images . For example , the source dataset may comprise a large number of images in various forms , such as embedded in documents . The images may comprise important information, such as flow charts , or non-inf ormative content , such as pictures that are intended only for improving the visual look of a document . The method 100 can extractsuch information from the source dataset to the grounding dataset .

[0048] The method 100 can, for example , improve the quality of the information used for machine learning model grounding .

[0049] The method 100 can, for example , save storage and / or computing resources . With the method 100 , images in the source dataset can be pre-proces sed so that the information comprised in the images can be more eas ily used for machine learning model grounding without needing to provide the images to the machine learning model . Thus , the machine learning model does not need to process the images , the images do not need to be provided to the machine learning model in a prompt , and unnecessary information does not need to be added to the grounding dataset .

[0050] Fig . 2 illustrates a flow chart representation of a method according to another embodiment .

[0051] According to an embodiment , the generating 102 the description of each image in the source dataset comprises generating the description of each image in the source dataset using an image description machine learning model .

[0052] The image description machine learning model may comprise any machine learning model config- ured / trained to generate descriptions of images .

[0053] According to an embodiment , the method 100 further comprises classifying 201 each image in the source dataset using an image classification machine learningmodel , thus obtaining the classification of each image in the source dataset .

[0054] The image classification machine learning model may comprise any machine learning model config- ured / trained to classify images .

[0055] According to an embodiment , the method 100 further comprises determining 202 the number of characters in each image in the source dataset using a character recognition machine learning model .

[0056] The character recognition machine learning model may comprise any machine learning model config- ured / trained to identify the number of characters in an image .

[0057] It should be appreciated that the order of operations illustrate in the embodiment of Fig . 2 is only exemplary and the operations can also be performed in various other orders . For example , operations 102 , 201 , and 202 may be performed in any order .

[0058] The image description machine learning model , the image classification machine learning model , and / or the character recognition machine learning model may be di fferent from the machine learning model used for the grounding .

[0059] In some embodiments , the image description machine learning model , the image classification machine learning model , and / or the character recognition machine learning model may be implemented using one machine learning model . For example , in some embodiments , the image description machine learning model , the imageclassification machine learning model, and the character recognition machine learning model may be implemented using an LLM. For example, an LLM may be provided a prompt requesting the LLM to provide a description of an image, to classify the image, and to identify the number of characters in the image.

[0060] For example, by providing the following prompt to the OpenAI GPT-4o model, GPT-4o can provide a description of an image, classify the image, and identify the number of characters in the image:"You are a helpful assistant that is tasked to extract information from images. Please describe what is in the attached image. Do not describe the size of format of the image, only the contents. Also classify the image to one of the following categories. Try to include in the description useful information such as categories in column charts or pie diagrams and the values.- Chart, the image contains a chart.- Diagram, the image contains a diagram- Flow chart, the image contains a flow chart- Text, the image contains mainly textual information- Multiple, the image contains more than one of the previously listed elements.- No info, the image does not contain text, charts, diagram, flow charts or other elements that contain useful information.- Other, the image does not belong to any of above mentioned categories.Please return a JSON-structure with the following four elements :- Name : Name of the image file- Character count : the number of characters found in the image- Category : The above described category of the image- Description : The description of the image in Finnish language . "

[0061] According to an embodiment , at least one image in the source dataset comprises a plurality of sub images ; the generating 102 the description of each image in the source dataset comprises generating a description of each sub image in the plurality of sub images ; and the generating 103 the grounding dataset by selecting descriptions of images in the source dataset to the grounding dataset comprises selecting descriptions of sub image in the plurality of sub images to the grounding dataset based at least on a clas sification of each sub image in the plurality of sub images and a number of characters in each sub image in the plurality of sub images .

[0062] Herein, a sub image may refer to an image with another image . For example , an image in the source dataset may comprise both a flow chart and an organi zational chart , wherein the f low chart and the organi zational chart can be considered as sub images of the image .

[0063] In some embodiments , the classification of an image can indicate that an image comprises a plurality of sub images . For example , with the prompt disclose above , images comprising sub images can be identified .

[0064] Any disclosure herein in relation to images in the source dataset may also apply to sub images .

[0065] Fig . 3 illustrates a flow chart representation of a method according to another embodiment .

[0066] In the embodiment of Fig . 3 , an image is obtained 301 , a description of the image is generated 302 , the number of characters N in the image is determined 303 , and the image is classified 304 .

[0067] According to an embodiment , the selecting descriptions of images in the source dataset to the grounding dataset based at least on the classification of each image in the source dataset and the number of characters in each image in the source dataset comprises selecting descriptions of images in the source dataset to the grounding dataset based on whether an image comprises more characters than a threshold number of characters .

[0068] For example , in the embodiment of Fig . 3 , in operation 305 , the number of characters in the image N can be compared to a threshold number of characters ^threshold - I f N > Nthresho ld fthe procedure can move to operation 306 . I f not , the procedure can move to operation 307 , where the description of the image is not selected to the grounding dataset . From operation 307 , the procedure can be move to operation 309 , where it is determined whether there are more images in the source dataset to be processed . I f there are more images , the procedure can move back to operation 301 so that the next image can be processed . I f there are no more images ,the procedure can move to operation 310 and the procedure can end.

[0069] The threshold number of characters Nthrestl0idcan be, for example, 3, 4, 5, 6, 7, 8, 9, 10, 15, or 20.

[0070] According to an embodiment, the selecting descriptions of images in the source dataset to the grounding dataset based at least on the classification of each image in the source dataset and the number of characters in each image in the source dataset comprises selecting descriptions of images in the source dataset to the grounding dataset based on whether an image belongs to a predetermined image class.

[0071] For example, in the embodiment of Fig. 3, in operation 306, it can be checked whether the image class of the image is informative. If the image class is informative, the description of the image can be selected to the grounding dataset in operation 308 and the procedure can then move to operation 309. If the image class is not informative, the procedure can move to operation 307.

[0072] According to an embodiment, the selecting descriptions of images in the source dataset to the grounding dataset based at least on a classification of each image in the source dataset and a number of characters in each image in the source dataset comprises, in response to an image comprising more characters than a threshold number of characters and the image belonging to a predetermined image class, selecting the description of the image to the grounding dataset.

[0073] For example , in the embodiment of Fig . 3 , a description of an image is selected to the grounding dataset if the number of characters in the image is greater that than the threshold number of characters and the image belongs to an informative class .

[0074] In some embodiments , the predetermined image class may comprise , for example , an informative image class indicating that the image comprises information . Such an informative image class may or may not comprise a plurality of other image classes .

[0075] Fig . 4 illustrates a schematic representation of image classes according to an embodiment .

[0076] According to an embodiment , the classification of each image in the source dataset classifies the image to at least one image class from a plurality of image classes , wherein the plurality of image classes comprises a non-inf ormative image class and at least one informative image class .

[0077] For example , in the embodiment of Fig . 4 , the plurality of image classes 400 comprises a non-informa- tive image class 401 and a plurality of informative image classes 402 . The plurality of informative image classes 402 can comprise a chart class 403 indicating that the image comprises a chart , a diagram class 404 indicating that the image comprises a diagram, a flow chart class 405 indicating that the image comprises a flow chart , a text class 406 indicating that the image comprises mainly or only text , a multiple sub imagesclass 407 indicating that the image comprises a plurality of sub images , and an "other" class 408 indicating that the image comprises some other content .

[0078] The non-inf ormative image class 401 may indicate that the image does not comprise any informative content . For example , the image may only comprise a picture that i s intended only for improving the visual appearance of a document .

[0079] In other embodiments , the plurality of image classes 400 may comprise other classes than what is depicted in the embodiment of Fig . 4 . For example , in some embodiments , the plurality of image classes 400 may comprise only the non-inf ormative image class 401 and an informative image class .

[0080] In some embodiments , the plurality of informative image classes 402 may be an image class and image classes 403 - 408 may be sub classes of the informative image class .

[0081] In other embodiments , the informative image class may be a separate image class from other image classes , such as image classes 403 - 408 illustrated in the embodiment of Fig . 4 . In such embodiments , any image belonging to the image classes 403 - 408 may also belong to the informative image clas s . Thus , an image may belong to more than one image class .

[0082] Fig . 5 illustrates a schematic representation of a computing device according to an embodiment .

[0083] According to an embodiment , a computing device 500 comprises at least one processor 501 and at leastone memory 502 including computer program code, the at least one memory 502 and the computer program code configured to, with the at least one processor 501, cause the computing device 500 to perform the method 100.

[0084] The computing device 500 may comprise at least one processor 501. The at least one processor 501 may comprise, for example, one or more of various processing devices, such as a co-processor, a microprocessor, a digital signal processor (DSP) , a processing circuitry with or without an accompanying DSP, or various other processing devices including integrated circuits such as, for example, an application specific integrated circuit (ASIC) , a field programmable gate array (FPGA) , a microprocessor unit (MCU) , a hardware accelerator, a special-purpose computer chip, or the like.

[0085] The computing device 500 may further comprise a memory 502. The memory 502 may be configured to store, for example, computer programs and the like. The memory 502 may comprise one or more volatile memory devices, one or more non-volatile memory devices, and / or a combination of one or more volatile memory devices and nonvolatile memory devices. For example, the memory 502 may be embodied as magnetic storage devices (such as hard disk drives, magnetic tapes, etc.) , optical magnetic storage devices, and semiconductor memories (such as mask ROM, PROM (programmable ROM) , EPROM (erasable PROM) , flash ROM, RAM (random access memory) , etc.) .

[0086] The computing device 500 may further comprise other components not illustrated in the embodiment ofFig . 5 . The computing device 500 may comprise , for example , an input / output bus for connecting the computing device 500 to other devices . Further, a user may control the computing device 500 via the input / output bus .

[0087] When the computing device 500 is configured to implement some functionality, some component and / or components of the computing device 500 , such as the at least one processor 501 and / or the memory 502 , may be configured to implement this functionality . Furthermore , when the at least one processor 501 is configured to implement some functionality, this functionality may be implemented using program code comprised, for example , in the memory 502 .

[0088] The computing device 500 may be implemented at least partially using, for example , a computer, some other computing device , or similar .

[0089] The method 100 and / or the computing device 500 may be utilised in, for example , chatbot applications . For example , a chatbot may be configured to answer questions presented by users via a chat interface . The chatbot may be configured to utili ze a machine learning model , such as an LLM, to answers the questions . The machine learning model can be grounded using the grounding dataset obtained using the method 100 .

[0090] The method 100 and / or the computing device 500 may be utilised in, for example , voicebot applications . A voicebot may be configured to obtain information from users by, for example , phone and convert the voice information into text information using automatic speechrecognition (ASR) . Thus , the voicebot can function similarly to a chatbot as disclosed above but us ing audio instead of a chat interface .

[0091] Any range or device value given herein may be extended or altered without losing the effect sought . Also any embodiment may be combined with another embodiment unless explicitly disallowed .

[0092] Although the subj ect matter has been described in language specific to structural features and / or acts , it is to be understood that the subj ect matter defined in the appended claims is not necessarily limited to the specific features or acts described above . Rather, the specific features and acts described above are disclosed as examples of implementing the claims and other equivalent features and acts are intended to be within the scope of the claims .

[0093] It will be understood that the benefits and advantages described above may relate to one embodiment or may relate to several embodiments . The embodiments are not limited to those that solve any or all of the stated problems or those that have any or all of the stated benefits and advantages . It wil l further be understood that reference to ' an ' item may refer to one or more of those items .

[0094] The steps of the methods described herein may be carried out in any suitable order, or simultaneously where appropriate . Additionally, individual blocks may be deleted from any of the methods without departingfrom the spirit and scope of the subj ect matter described herein . Aspects of any of the embodiments described above may be combined with aspects of any of the other embodiments described to form further embodiments without losing the effect sought .

[0095] The term ' comprising ' is used herein to mean including the method, blocks or elements identified, but that such blocks or elements do not comprise an exclusive list and a method or apparatus may contain additional blocks or elements .

[0096] It will be understood that the above description is given by way of example only and that various modif ications may be made by those ski lled in the art . The above specification, examples and data provide a complete description of the structure and use of exemplary embodiments . Although various embodiments have been described above with a certain degree of particularity, or with reference to one or more individual embodiments , those skilled in the art could make numerous alterations to the disclosed embodiments without departing from the spirit or scope of this specification .

Claims

CLAIMS :

1. A computer-implemented method (100) for grounding a machine learning model, the method (100) comprising : obtaining (101) a source dataset, wherein the source dataset comprises a plurality of images; generating (102) a description of each image in the source dataset, wherein the generating (102) the description of each image in the source dataset comprises generating the description of each image in the source dataset using an image description machine learning model; generating (103) a grounding dataset by selecting descriptions of images in the source dataset to the grounding dataset based at least on a classification of each image in the source dataset and a number of characters in each image in the source dataset; and using (104) the grounding dataset for grounding a machine learning model.

2. The computer-implemented method (100) according to claim 1, wherein the method (100) further comprises classifying each image in the source dataset using an image classification machine learning model, thus obtaining the classification of each image in the source dataset.3 . The computer-implemented method ( 100 ) according to any preceding claim, wherein the method ( 100 ) further comprises determining the number of characters in each image in the source dataset using a character recognition machine learning model .4 . The computer-implemented method ( 100 ) according to any preceding claim, wherein the classification of each image in the source dataset classifies the image to at least one image class from a plurality of image classes ( 400 ) , wherein the plurality of image classes ( 400 ) comprises a non-inf ormative image class ( 401 ) and at least one informative image class ( 402 ) .5 . The computer-implemented method ( 100 ) according to any preceding claim, wherein the selecting descriptions of images in the source dataset to the grounding dataset based at least on the classification of each image in the source dataset and the number of characters in each image in the source dataset comprises selecting descriptions of images in the source dataset to the grounding dataset based on whether an image comprises more characters than a threshold number of characters .6 . The computer-implemented method ( 100 ) according to any preceding claim, wherein the selecting descriptions of images in the source dataset to the grounding dataset based at least on the classification of each image in the source dataset and the number ofcharacters in each image in the source dataset comprises selecting descriptions of images in the source dataset to the grounding dataset based on whether an image belongs to a predetermined image class .7 . The computer-implemented method ( 100 ) according to any preceding claim, wherein the selecting descriptions of images in the source dataset to the grounding dataset based at least on a classification of each image in the source dataset and a number of characters in each image in the source dataset comprises , in response to an image comprising more characters than a threshold number of characters and the image belonging to a predetermined image class , selecting the description of the image to the grounding dataset .8 . The computer-implemented method ( 100 ) according to any preceding claim, wherein the using the grounding dataset for grounding the machine learning model comprises providing a prompt to the machine learning model , wherein the prompt comprises a request and instructions to the machine learning model to respond to the request using the grounding dataset .9 . The computer-implemented method ( 100 ) according to any preceding claim, wherein at least one image in the source dataset comprises a plurality of sub images ;the generating (102) the description of each image in the source dataset comprises generating a description of each sub image in the plurality of sub images; and the generating (103) the grounding dataset by selecting descriptions of images in the source dataset to the grounding dataset comprises selecting descriptions of sub image in the plurality of sub images to the grounding dataset based at least on a classification of each sub image in the plurality of sub images and a number of characters in each sub image in the plurality of sub images.

10. The computer-implemented method (100) according to any preceding claim, wherein the machine learning model comprises a large language model.

11. A computing device (500) , comprising at least one processor (501) and at least one memory (502) including computer program code, the at least one memory (502) and the computer program code configured to, with the at least one processor (501) , cause the computing device (500) to perform the method (100) according to any preceding claim.

12. A computer program product comprising program code configured to perform the method (100) according to any of claims 1 - 10 when the computer program product is executed on a computer.

Images