High-speed and efficient classification system

Abstract

To provide a system and a method for generating a trained classification model, for classifying an entity, or for classifying the similarity of an entity and another entity.SOLUTION: A cloud system comprises a database for storing the characteristics of a plurality of first entities. Each of the plurality of first entities is classified according to a first classification. The cloud system includes a training module for training a classification model, in order to classify an input entity using the first classification, or to classify the similarity of an input entity and the plurality of first entities, on the basis of the characteristics of the input entity. The classification model uses the characteristics of the plurality of first entities as training data in order to be trained. The cloud system further includes a transmitter for transmitting the trained classification model to a client device, for use in the client device.SELECTED DRAWING: Figure 4

Description

Content of disclosure

[0001] [Technical field] The present disclosure relates to systems, computer-implemented methods, and computer programs for classifying entities or classifying the similarity of an entity to other entities.

[0002] 〔background〕 The number of devices connected to the Internet as part of the Internet of Things (IoT) is expected to grow exponentially in the coming years. These devices are intended to collect data and send that data to "the cloud." As the number of devices and the amount of data they collect increases significantly, the demand for computation and storage on cloud servers becomes an issue. In addition, there is a finite bandwidth issue, which means that sending data to the cloud can also be cumbersome.

[0003] The number of applications of machine learning is also growing rapidly. It is becoming a common feature of many platforms and solutions and is particularly useful for processing large amounts of data for pattern recognition, anomaly detection, and forming data joins.

[0004] Electronic devices, particularly those wirelessly connected to the Internet, including mobile phones, consume a significant portion of their energy usage to connect and transfer data to and from the Internet via Wi-Fi, telecommunications data networks or other means. In devices such as mobile phones, the energy stored in the battery becomes a constraint on the usage time of the device when data needs to be sent to and from the Internet frequently and for long periods of time. In general, cloud processing and analytics platforms, including those with machine learning capabilities, often require the device to keep communicating with the cloud periodically so that data stored in the cloud is kept up to date with data collected on the device. Thus, from a usage perspective, it is undesirable for cloud-based machine learning systems to often require large amounts of frequent data transmission.

[0005] Classifying or "segmenting" entities (such as devices or users) based on data associated with these entities (such as their characteristics) is a powerful means of identifying specific entities of interest for specific reasons. For example, it may be desirable to be able to classify a particular device in a system as faulty, so that necessary actions can be taken to return the system to a properly functioning state. In another example, it may be desirable to be able to classify a device or user as representing a security threat, so that appropriate preventative measures can be taken to neutralize the threat.

[0006] Categorizing entities using set criteria may be restrictive and may result in entities being missing from a particular classification, for example, some entities that do not meet the entire criteria may still be relevant enough that they should also be included in a classification group or segment.

[0007] Automating the decision regarding which entities to give a particular classification is especially useful for systems that need to classify entities in near real-time. Furthermore, it would take a very long time to manually classify entities and for such systems to remain scalable for large data sets. For a related set of entities with a particular classification, the decision regarding whether to include a user or device in the class should be based on relevance or similarity to entities that meet the classification criteria. For example, there may be unknown correlations between the characteristics of the entities, which may result in an entity being missing from a particular classification.

[0008] Machine learning algorithms can be trained on datasets that continually change as more data is uploaded and received. Retraining involves learning new parameters for the model, and these new model parameters may render classifications previously assigned to entities obsolete. However, reclassifying all previously classified entities can be computationally expensive and time-constrained. For example, in a typical cloud environment, a system with 100 million classified entities would reclassify all 100 million entities for a given classification with new model parameters.

[0009] In view of the above, there is a need for a system that can quickly and accurately classify entities and identify entities that are similar to entities of a particular classification. This should desirably be accomplished in a manner that efficiently utilizes processing resources and bandwidth. Additionally, there is a need to reclassify previously classified entities in a computationally efficient manner.

[0010] 〔overview〕 In one aspect of the present invention, there is provided a system for classifying an entity or classifying an entity's similarity to other entities, the system comprising: The server includes: a storage means configured to store characteristics of a first plurality of entities, each of the first plurality of entities being classified according to a first classification; a training module configured to train a classification model to classify the input entity with a first classification or to classify an similarity between the input entity and a first plurality of entities based on characteristics of the input entity, the classification model being trained using the characteristics of the first plurality of entities as training data; a transmitter configured to transmit the trained classification model to a client device; Including, The system further includes a client device, the client device comprising: a receiver configured to receive the trained classification model; a memory configured to store the trained classification model; and a processor configured to process data comprising a characteristic of at least one input entity using a trained classification model stored on a client device and cause the trained classification model to output a signal, the signal comprising: classifying at least one input entity with a first classification, or A processor for classifying an affinity between the at least one input entity and the first plurality of entities.

[0011] In the present invention, a machine learning (ML) model (i.e., a classification model) is trained using characteristics of an entity with a first classification to classify another entity with the same classification. In this way, unclassified entities can be classified more accurately than using a limited set of pre-determined criteria. Furthermore, the ML model is trained to classify the similarity of an entity to an entity with a first classification. This allows for identification of entities that are similar to other classified entities.

[0012] In the present invention, the ML model is trained on a server in the cloud. Then, the ML model is distributed to the client devices, where the ML model is executed to output the classification. This reduces the burden on the cloud server and the transmission network. This also reduces the power consumption of the client device, since the client device is not required to train the ML model. Furthermore, once the model parameters are received at the client device on the edge, the client device can continue to classify and react to the data being generated on the device without needing a connection to the Internet. This makes the device classification autonomous and robust, which is useful in the face of variable or unstable connectivity and failure of the central server.

[0013] Processing of data related to the input entity is done on the processor of the client device without the need to send this data to the cloud. This eliminates the need for constant and frequent communication between the cloud server and the client device. Therefore, the solution is less energy intensive, which is especially useful for battery-powered devices. It also helps to prevent sensitive data from being sent to malicious entities, improving security and data privacy.

[0014] A further advantage of the present invention is that the number of processors, and therefore the processing power of the system, is proportional to the number of devices involved. Also, as the amount of data transmitted is reduced, bandwidth constraints become less of an issue. Thus, the system is inherently scalable.

[0015] Furthermore, the time it takes to process the data of a device remains constant regardless of the size of the system. From the perspective of the client device or user, processing can occur in near real-time regardless of the size of the network. In the context of a data management system, having negligible latency between data collection and processed output is highly advantageous as it allows for live interaction with or reaction to the device or user.

[0016] In summary, the benefits of the proposed system are improved power consumption for the end user, continuous real-time processing, faster response to events, more efficient use of processing resources, the ability to process private, sensitive, and / or personal data locally rather than sending it to the cloud, and increased resilience to connectivity issues.

[0017] In another aspect of the invention, there is a server for generating a trained classification model for classifying an entity or classifying an similarity of an entity to other entities, the server comprising: a storage means configured to store characteristics of a first plurality of entities, each of the first plurality of entities being classified according to a first classification; a training module configured to train a classification model to classify the input entity with a first classification or to classify a similarity between the input entity and a first plurality of entities based on characteristics of the input entity, the classification model being trained using the characteristics of the first plurality of entities as training data; Including, The server The apparatus further includes a transmitter configured to transmit the trained classification model to a client device for use by the client device.

[0018] In another aspect of the invention, there is a client device for classifying an entity or classifying an affinity between an entity and other entities, the client device comprising: a receiver configured to receive a trained classification model from a server; The trained classification model is trained by the server using characteristics of the first plurality of entities classified with the first classification as training data to classify the input entity with a first classification or classify a similarity between the input entity and the first plurality of entities based on characteristics of the input entity; The client device is a memory configured to store the trained classification model; and a processor configured to process data comprising a characteristic of at least one input entity using a trained classification model stored on a client device and cause the trained classification model to output a signal, the signal comprising: classifying at least one input entity with a first classification, or and a processor for classifying an affinity between the at least one input entity and the first plurality of entities.

[0019] In another aspect of the invention, there is a computer-implemented method for classifying an entity or classifying an entity's similarity to other entities, the method comprising: obtaining, at the server, characteristics of a first plurality of entities, each of the first plurality of entities being classified with a first taxonomy; training a classification model at the server to classify the input entity with a first classification or classify a similarity between the input entity and a first plurality of entities based on characteristics of the input entity, the classification model being trained using the characteristics of the first plurality of entities as training data; transmitting the trained classification model from the server to the client device; storing the trained classification model at the client device; using the trained classification model stored at the client device to process data comprising a characteristic of at least one input entity and causing the trained classification model to output a signal, the signal comprising: classifying at least one input entity with a first classification, or classifying an affinity between the at least one input entity and the first plurality of entities.

[0020] In another aspect of the invention, there is a computer-implemented method for generating a trained classification model for classifying an entity or classifying an entity's similarity to other entities, the method comprising: obtaining, at the server, characteristics of a first plurality of entities, each of the first plurality of entities being classified with a first taxonomy; training a classification model at the server to classify the input entity with a first classification or classify a similarity between the input entity and a first plurality of entities based on characteristics of the input entity, the classification model being trained using the characteristics of the first plurality of entities as training data; transmitting the trained classification model from the server to the client device for use by the client device; Includes.

[0021] In another aspect of the invention, there is a computer-implemented method for classifying an entity or classifying an entity's similarity to other entities, the method comprising: receiving, at the client device, the trained classification model from the server; The trained classification model is trained by the server using characteristics of the first plurality of entities classified with the first classification as training data to classify the input entity with the first classification or classify a similarity between the input entity and the first plurality of entities based on the characteristics of the input entity; storing the trained classification model at the client device; using the trained classification model stored at the client device to process data comprising a characteristic of at least one input entity and causing the trained classification model to output a signal, the signal comprising: classifying at least one input entity with a first classification, or classifying an affinity between at least one input entity and a first plurality of entities; Includes.

[0022] The trained classification model may be configured to output a similarity score to classify the similarity between the input entity and the first plurality of entities.

[0023] A receiver at the client device may be configured to receive the threshold similarity score. A processor may be configured to perform an action if the similarity score of the at least one input entity satisfies the threshold similarity score.

[0024] The computer-implemented method can include receiving a threshold similarity score and can include performing an action if the similarity score of the at least one input entity meets the threshold similarity score.

[0025] The processor may be configured to perform a plurality of different actions at the client device, each of the actions being performed in response to a different similarity score being associated with the at least one input entity.

[0026] The computer-implemented method can include performing a plurality of different actions, each of which is performed in response to a different similarity score being associated with at least one of the input entities.

[0027] The receiver at the client device may be configured to receive the threshold similarity score, and the processor is configured to classify the at least one input entity with the first classification if the similarity score of the at least one input entity satisfies the threshold similarity score.

[0028] The computer-implemented method may include receiving a threshold similarity score and classifying the at least one input entity with a first classification if the similarity score of the at least one input entity meets the threshold similarity score.

[0029] The processor may be configured to perform an action in response to the signal.

[0030] The computer-implemented method can include performing an action in response to the signal.

[0031] The actions to be performed in response to a signal may be defined by an operator of the system.

[0032] The actions performed in response to the signal may include generating content instructions at the client device to display content, the content being based on a classification of the at least one input entity with a first classification or based on a similarity of the at least one input entity to a first plurality of entities.

[0033] The action may include sending a message indicating that the at least one input entity has been classified with a first classification or indicating a similarity of the at least one input entity to a first plurality of entities.

[0034] The receiver may be configured to obtain, at the server, characteristics of a second plurality of entities distinct from the first plurality of entities. The server may include a model analyzer configured to process data including the characteristics of the second plurality of entities using the trained classification model to output a similarity score for each of the second plurality of entities, each similarity score output by the trained classification model being associated with one of the second plurality of entities and classifying a similarity between the associated entity and the first plurality of entities.

[0035] The computer-implemented method may include obtaining characteristics of a second plurality of entities that differ from the first plurality of entities. The computer-implemented method may include processing data including the characteristics of the second plurality of entities using the trained classification model to output a similarity score for each of the second plurality of entities, each similarity score output by the trained classification model being associated with one of the second plurality of entities and classifying a similarity between the associated entity and the first plurality of entities.

[0036] The model analyzer may be configured to calculate a number of second entities associated with each similarity score output by the trained classification model.

[0037] The computer-implemented method can include calculating a number of second entities associated with each similarity score output by the trained classification model.

[0038] The model analyzer may be configured to calculate a number of second entities having similarity scores that satisfy a number of different thresholds.

[0039] The computer-implemented method can include calculating a number of second entities that have similarity scores that meet a number of different thresholds.

[0040] The model analyzer may be configured to output a graphical display showing the number of second entities associated with each similarity score output by the trained classification model.

[0041] The computer-implemented method can include outputting a graphical display showing the number of second entities associated with each similarity score output by the trained classification model.

[0042] The model analyzer may be configured to output a graphical display indicating the number of second entities having similarity scores that meet a number of different thresholds.

[0043] The computer-implemented method can include outputting a graphical display indicating the number of second entities having similarity scores that meet a number of different thresholds.

[0044] A receiver at the client device may be configured to receive a threshold similarity score set based on the number of second entities associated with each similarity score output.

[0045] The computer-implemented method can include receiving a threshold similarity score set based on a number of second entities associated with each similarity score output.

[0046] A receiver at the client device may be configured to receive a threshold similarity score that is set based on a number of second entities having similarity scores that meet a number of different thresholds.

[0047] The computer-implemented method can include receiving a threshold similarity score that is set based on a number of second entities having similarity scores that meet a number of different thresholds.

[0048] A first plurality of entities classified with the first classification may satisfy a first set of predetermined criteria. The predetermined criteria may be adjustable.

[0049] The system may further include an operator device, the operator device comprising: an interface configured to allow an operator to specify a first set of criteria for comparison with characteristics of entities, where entities that meet the first set of criteria are classified with a first classification; and a transmitter configured to transmit a first set of criteria to a server; Including, The server further includes a receiver configured to receive the first set of criteria; The training module is configured to compare a first set of criteria to characteristics of the plurality of entities and classify entities that satisfy the first set of criteria with a first classification.

[0050] The computer-implemented method can include receiving a first set of criteria specified by an operator for comparison to characteristics of the entities, and entities that satisfy the first set of criteria are classified with a first classification.

[0051] The computer-implemented method can include transmitting a first set of criteria to a server.

[0052] The computer-implemented method can include receiving a first set of criteria.

[0053] The computer-implemented method can include comparing a first set of criteria against characteristics of a plurality of entities and classifying entities that satisfy the first set of criteria with a first classification.

[0054] The training module can be configured to calculate a number of entities that meet a first set of criteria.

[0055] The computer-implemented method can include calculating a number of entities that satisfy a first set of criteria.

[0056] The training module can be configured to initiate training of the classification model based on the number of entities that satisfy the first set of criteria.

[0057] The computer-implemented method may include initiating training of a classification model based on a number of entities that satisfy a first set of criteria.

[0058] The client device can include a transmitter configured to transmit a web element request indicating the web element.

[0059] The computer-implemented method can include sending a web element request indicating the web element.

[0060] The system may further include a web element server configured to transmit the web elements and the trained classifier to the client device in response to a web element request.

[0061] The computer-implemented method can include transmitting the web element and the trained classifier to the client device in response to the web element request.

[0062] The entity may be a computing device, and thus the characteristics may describe the capabilities of the computing device.

[0063] A computing device having a first classification may be classified as having failed or potentially failing.

[0064] The entity may be a user, and the characteristics may describe attributes of the user and / or activities of the user.

[0065] A user having a first classification may be classified as a potential security threat.

[0066] The trained classifier may be configured to receive characteristics of an entity as input.

[0067] Training the classification model may include determining a weighting for each of a plurality of parameters, the weighting associated with each parameter indicating the amount of influence the parameter has on the output signal.

[0068] Optionally, the computer-implemented method includes determining whether a weighting associated with at least one of the parameters meets a threshold weighting. The training module may be configured to determine whether a weighting associated with at least one of the parameters meets a threshold weighting. Optionally, at least one parameter associated with a weighting that does not meet (or is below) the threshold weighting is not used by the trained classifier in classifying the entity or in classifying the similarity of the at least one input entity to the first plurality of entities. Optionally, the identified parameter is not used because the parameter is not included in the processing performed by the trained classification model. Optionally, at least one parameter associated with a weighting that does not meet (or is below) the threshold is omitted from the trained model. The training module at the server may perform the omission of at least one parameter from the trained model. Optionally, at least one parameter is omitted because it is deleted from the model or the trained model is sent to the client device without the parameter (or parameters) that are omitted. The threshold weighting may be based on the weighting of the parameters. For example, the predetermined threshold may be an average of the weighting of the parameters. The threshold weighting may be adjustable, and optionally the threshold weighting is adjustable by the user.

[0069] Optionally, the parameters associated with the lowest weightings are not used by the trained classifier in classifying the entities or in classifying the similarity between the at least one input entity and the first plurality of entities.

[0070] Optionally, a predetermined number of parameters associated with the lowest weightings are not used by the trained classifier in classifying the entities or in classifying the similarity between the at least one input entity and the first plurality of entities.

[0071] The computer-implemented method may include retraining the classification model to classify the input entity with the first classification or to classify the similarity of the input entity to the first plurality of entities. Optionally, the training module is configured to retrain the classification model to classify the input entity with the first classification or to classify the similarity of the input entity to the first plurality of entities. The classification model may be retrained using a second, different set of characteristics of the plurality of entities classified with the first classification as training data. The classification model may be retrained using a second, different set of characteristics of the plurality of entities classified with the first classification that are different from the first plurality of entities as training data.

[0072] The computer-implemented method may include retraining the classification model at a predetermined frequency or after a predetermined time interval. The training module may be configured to retrain the classification model at a predetermined frequency or after a predetermined time interval. The predetermined frequency or the predetermined time interval may be adjustable, and optionally the predetermined frequency is adjustable by a user.

[0073] The computer-implemented method may include transmitting the retrained classification model to the client device. The transmitter may be configured to transmit the retrained classification model to the client device.

[0074] The computer-implemented method may include receiving a retrained classification model at a client device, storing the retrained classification model at the client device, and processing data including characteristics of at least one input entity using the retrained classification model stored at the client device to cause the retrained classification model to output a signal, the signal classifying the at least one input entity with a first classification or classifying a similarity of the at least one input entity to a first plurality of entities.

[0075] The receiver in the client device may be configured to receive the retrained classification model. The memory may be configured to store the retrained classification model. The processor may be configured to use the retrained classification model stored on the client device to process data including characteristics of the at least one input entity to cause the retrained classification model to output a signal, the signal classifying the at least one input entity with a first classification or classifying a similarity of the at least one input entity to a first plurality of entities.

[0076] The computer-implemented method may include transmitting the retrained classification model to the client device by only transmitting parameters of the retrained model that differ from parameters of the trained model. The transmitter may be configured to transmit the retrained classification model to the client device by only transmitting parameters of the retrained model that differ from parameters of the trained model.

[0077] The computer-implemented method may include determining whether the input entity is classified with an initial classification, and processing data including at least one characteristic of the input entity using the stored classification model only if the stored classification model differs from the classification model used to classify the input entity with the initial classification.

[0078] A processor at the client device may be configured to determine whether the input entity is classified with the initial classification, and the processor is configured to process data including at least one characteristic of the input entity using the stored classification model only if the stored classification model differs from the classification model used to classify the input entity with the initial classification.

[0079] The computer-implemented method may include determining whether the input entity is classified with the initial classification, and obtaining a retrained classification at the client device from the server only if a most recently trained classification model at the server differs from the classification model used to classify the input entity with the initial classification, and processing data including at least one characteristic of the input entity using the retrained classification model.

[0080] A processor at the client device may be configured to determine whether the input entity is classified with the initial classification, and to obtain a retrained classification from the server only if the most recently trained classification model at the server differs from the classification model used to classify the input entity with the initial classification, and to process data including at least one characteristic of the input entity using the retrained classification model.

[0081] The trained classification model can include a linear regression model.

[0082] In another aspect of the invention there is a computer program comprising instructions which, when executed by a computer, cause the computer to carry out the computer-implemented methods described herein.

[0083] In another aspect of the invention, there is a computer-readable medium containing instructions that, when executed by a computer, cause the computer to perform the computer-implemented methods described herein.

[0084] In another aspect of the invention there is a data carrier signal carrying a computer program as described herein.

[0085] Embodiments of the invention will now be described, by way of example only, with reference to the following drawings, in which: [Brief description of the drawings]

[0086] [Figure 1] SUMMARY OF THE DISCLOSURE An overview of a system for classifying entities or classifying their similarity to other entities is presented. [Diagram 2] 1 shows a flowchart illustrating steps in a method for classifying an entity or classifying its similarity to other entities. [Diagram 3] 1 shows a graphical representation of the output from a trained classification model. [Figure 4] Further details of the system are given. [Diagram 5] We outline how to classify entities using a trained classification model. [Figure 6] 1 illustrates an example of an electronic device of the system.

[0087] Detailed Description 1, there is a system 1 that includes a cloud network 3 that collates data describing entities from a third party data provider 1. The cloud network 3 inputs the data into a classification database 7 that is used to train machine learning models via a machine learning service 2 in a server. The cloud 3 also supplies information to and receives data from a cloud online dashboard 4. The cloud network 3 may be, for example, the Internet.

[0088] The machine learning service 2 outputs the model parameters to a cloud network 3. The model parameters are injected into code segments or software development kits (SDKs) that are sent through a content delivery network 5 to client devices 6 where the segments are processed. The client devices 6 may be referred to as edge devices in the system 10. Each client device 6 may take the form of a mobile phone, a tablet, a laptop computer, a computer, or another form of electronic device that includes processing means and means for communicating data to and from the Internet.

[0089] Transmit first party data from the edge devices 6 to the cloud network 3. These data may indicate characteristics of the entity, such as characteristics of the edge device, characteristics of the hardware within the edge device, or characteristics describing attributes of a user of the edge device 6. The data transmitted from the edge devices 6 can take many forms, such as, but not limited to, personal data and statistical data. These data are stored in a classification database 7.

[0090] An operator of the system 10 may wish to analyze entities or react to a particular portion of a set of entities. For example, a website owner may wish to analyze users or react to a particular type of user visiting the website. The type of entity detected by the system may be based on set criteria (or predetermined criteria) and the similarity of the entity to entities that meet the set criteria. The classification of entities is desirably done in real-time (or as near real-time as possible) so that reactions to entities having a particular classification can be made quickly.

[0091] 2, there is a method in which an operator defines a classification and initiates training of a classification model at a server. An operator can execute this method at an operator device 8 that interacts with a cloud online dashboard 4 and a machine learning service 2. The online dashboard 4 can display visual and statistical analysis in addition to having functionality to modify aspects of the system 10, including the creation and modification of classes (or "segments").

[0092] In this embodiment, the online dashboard 4 is accessed over the internet using a browser on a mobile phone, tablet, laptop computer, computer, or other internet connected device. In this manner, the online dashboard 4 displays the number of entities currently within the classifications defined by the operator and provides the operator with a means to initiate a machine learning (ML) similarity model, or in other words, to train a classification model. One means for the operator to interact with the online dashboard is preferably a digital button or slider that the operator clicks or drags with a cursor or other means, for example via a touch screen.

[0093] In step 100, an operator creates a classification (or "segment") by defining set criteria that an entity's characteristics must meet to be included in the segment. Depending on the data fields, the criteria may take the form of Boolean logic or may require more complex logic. Boolean logic may also be required to combine criteria. For example, a device may be classified as possibly failing if the device's temperature meets a certain threshold and the device has been operating for a certain length of time. In another example, a user 1 of a client device 6 may be classified with a particular classification if the user has no children and is under the age of 30.

[0094] Once the criteria for classification are defined in step 101, the operator device 8 evaluates the number of entities that meet the criteria. There may be a minimum or threshold number of entities that will be classified according to the set criteria. If the classification result does not meet this threshold, the method proceeds to step 102. If the threshold is met, the method proceeds to step 105.

[0095] In step 102, a classification model is trained using the training data to classify entities with classifications corresponding to the criteria set in step 100 or to classify the similarity of entities to entities that meet the criteria set in step 100. The trained classification model takes characteristics of entities as its input and classifies the entities or their similarity based on the input characteristics. The classification model is trained using characteristics of entities that meet the criteria set in step 100 as training data.

[0096] The classification model is trained with data characteristic of entities stored in the classification database 7. In step 103, once the classification model is trained, it is used to calculate the similarity of entities based on the stored data describing these entities. The level of similarity between each entity and entities that meet the pre-set criteria is output as a similarity metric, which may include a percentage. For example, if an entity is given a similarity metric of 99%, this indicates that the entity is nearly identical to entities that meet the pre-set criteria (at least with respect to the criteria being evaluated). On the other hand, if an entity is given a similarity metric of 1%, this indicates that the entity is very different from entities that meet the pre-set criteria (at least with respect to the criteria being evaluated).

[0097] The trained classification model is then used to estimate the number of entities in the classification for several different similarity metric values. Once the computation is complete, which may take several hours, the results can be output in graphical form in a similarity graph, an example of which is shown in Figure 3.

[0098] 3 shows a graphical display displayed on the operator device 8. The display shows the number of entities (whose data are stored in the classification database 7) that have been assigned a similarity metric that meets a particular threshold (i.e., 90%, 55%, 30%, etc.). In other words, the graph shows the number of entities that have a similarity score that is equal to or greater than a particular threshold.

[0099] As shown in FIG. 3, the number of entities that meet the threshold is inversely proportional to the similarity threshold. As the number of entities decreases, the similarity threshold increases. On the other hand, as the number of entities increases, the similarity threshold decreases. The similarity threshold can take values ​​between 0% and 100%. A similarity score of 100% results in the same number of entities being included in the segment as was achieved with the set criteria before training the classification model. By reducing the similarity score below 100%, some entities that did not meet the set criteria but whose data are very similar to that of the entities that met the criteria are added and given the same classification as the entities that met the set criteria. A lower similarity threshold results in more entities being included in the classification, although they are less similar to the original set of entities. The operator can choose how strict or loose the classification will be by appropriately selecting the similarity threshold. The operator can choose the number of entities, as shown on the y-axis of the graph, and the corresponding similarity can be selected accordingly.

[0100] In step 105, the operator specifies the preferred number of entities to include in the classification, or a desired similarity threshold, and then specifies what reaction should occur when an entity is classified. A reaction can be defined as a process by which entities of a particular classification can be passed on to other platforms.

[0101] The system will function with a minimum number of criteria specified by the operator, however, it is recommended that at least 10 criteria be specified for each classification in order for similarity model similarity scoring to perform to a sufficient level of accuracy.

[0102] In another embodiment, an algorithm can be used to initiate different actions depending on the similarity percentage assigned to an entity, which can be used instead of or in combination with a similarity threshold.

[0103] For example, if a user has a similarity score of 71%, the algorithm will show the user a general message. If the user has a similarity score of 89%, a more detailed message will be displayed, and if the user has a similarity score of 93%, an even more detailed message will appear. Those skilled in the art will appreciate that this system is broad in its applications and potential uses, and the example provided is only one of many.

[0104] In the following, with reference to FIG. 4, we explain the process and data flow that occurs when employing a machine learning similarity model to expand the number of entities included in a particular classification.

[0105] Data from the client device 200 and the third party source 201 are sent to and stored in a centralized database 202 in a cloud system 205. The data from the client device 200 and the third party source describe characteristics of an entity, such as a device or a user of the device. When an operator requests the execution of a machine learning model via an online dashboard 204, the data from the client device 200 and the third party source 201 is used to train a similar machine learning model 203 using a training module 203 in the system 205. The data for this graphical display described with reference to FIG. 3 is generated by a model analyzer 210.

[0106] Each measure of an entity is known as a feature of the model. Each feature is treated as a separate dimension in the model. In a preferred embodiment, the model is based on a logistic regression fit. However, it will be appreciated that many other alternative models may be used, such as linear regression or neural networks, with different mathematical properties suitable for different embodiments of the invention.

[0107] Training a model results in weighting of the model's features. As part of the training process, a regularizer is used to reduce the number of parameters (i.e., "features") used in the final model, and a penalty (regularization term) is applied to reduce the model size, which has the effect of removing model parameters with low weights (close to 0) that do not significantly affect the similarity score during model evaluation. Features with low weights cannot be used, as the trained classification model does not use them in determining similarity scores or classifying entities.

[0108] The training module 203 in the cloud system 205 can perform a regularization process that causes low weighted parameters that have little impact on the classification to be removed from the final set of model parameters. For example, parameters with weights below a predetermined threshold can be removed from the model. Since the parameters must be transmitted to the edge device, stored, and used during classification, a smaller number of parameters, which means fewer parameters transmitted to the edge device, can improve performance. In a specific embodiment, the dashboard 204 is configured to allow the user to adjust the level of regularization, for example, by adjusting pre-determined thresholds for parameter weights. This can achieve better device and network performance at the expense of a potential loss in classification accuracy. The system can provide a default level of regularization, such as a default threshold for weights. In this way, the user can select whether manual adjustment of the regularization level is required or not.

[0109] Once model training with normalization is complete, the model output parameters are packaged into a code portion 208 (or SDK) that is sent to and stored on the client device 6. In a preferred embodiment, the SDK contains generic SDK elements 209 and injected JavaScript segments 206 along with injected machine learning model parameters 207. The SDK 208 is sent to the client edge device 200 when the client edge device 200 accesses an operator specified website, such as a website for which the operator is the administrator.

[0110] 5, the trained classification model is executed at the client device 309 in response to an event to classify an entity. An event can be any form of interaction from an entity, such as a device or a user. This can be, for example, a user visiting a web page, selecting a button, or a user scrolling down a web page.

[0111] e x Events, shown at 300, are fed to an SDK on a client device 309. The events are passed to a classification engine 301, which aggregates data about the user from both first-party sources and third-party data providers and aggregators together. The classification engine 301 takes a first-party classification state 302 and a third-party classification state 303 and outputs an updated classification state 304. The output is represented as a vector of Boolean values. The classification state is input to a parameterized machine learning model 306 (i.e., a trained classification model), which receives parameters for the model from a cloud server via a content delivery network 305, as described above.

[0112] In a preferred embodiment, processing of the model involves a dot product between the segment state vector and the model parameters, followed by application to a sigmoid function to generate probabilities, however, it will be appreciated that alternative mathematical methods may be used to calculate the model outputs.

[0113] Segments are processed and calculated for each event that occurs. However, model parameter updates are much less frequent; for example, the parameters of the model at the client device may only change when the model in the cloud is retrained. Instead of sending all parameters of the retrained model, including parameters that have not changed, the cloud system 205 may send only the parameters that have changed when updating the model at the client device 309. This allows for more efficient use of bandwidth.

[0114] It is desirable to retrain the model as new data is received and uploaded. However, continually retraining the model is computationally expensive. However, in the system described herein, the model may be retrained at the server at a predetermined frequency. This predetermined frequency may be set to maintain a balance between model accuracy and processing resource usage. The predetermined frequency may be user-defined.

[0115] Once a model is trained, some of the classifications previously assigned to entities may not match the classifications output by the new model. It is therefore desirable to reclassify entities once the model is retrained. However, reclassifying entities can still be computationally expensive. For example, if 100 million previously classified entities are to be reclassified with a retrained model, this places a large processing load on the system. The system disclosed herein improves the efficiency of this process by making new model parameters available to the client device when classification is required on the device. The new model parameters may be sent to the client device when a previously classified entity is detected for reclassification, for example, when a previously classified user visits a web page. When a previously classified entity is detected, the new model parameters are pulled to the device for use. In this sense, reclassification is performed only when requested (e.g., when only 1M of 100M entities require classification), while distributing the computational load to the devices, making it more rapid and reducing strain on centralized cloud infrastructure.

[0116] The machine learning model outputs a similarity metric 307, expressed for example in the form of a percentage after a sigmoid function has been applied. The similarity is treated like any other state and is fed back to the classification engine 308 where it is updated after an event. The SDK sent to the client device 309 contains a similarity threshold (otherwise known as "precision") against which the output similarity classification is compared. The result is a boolean classification state of the machine learning model that is either TRUE or FALSE. TRUE means that the user is included in a segment based on their similarity score and FALSE means the opposite.

[0117] Once processing is complete and the user is presumed to be within the segment, a reaction may be triggered. Similarly, in another embodiment, it is contemplated to generate a reaction such that a reaction is triggered if the user does not enter the segment.

[0118] In this method, first-party and third-party entity data fields are inputs to a machine learning model. Including a larger number of fields in the model increases the likelihood of a more accurate similarity score, as dependencies between all fields can be evaluated and accounted for. As part of the model training, weights are derived for each feature. The weights indicate the degree of impact that any given feature has on segment size. Features with very low weights, indicating a small impact on segment size, are discarded. Various methods for determining which features to remove can be used. These include, but are not limited to: A threshold weighting set at a certain predefined value. Features with weighting less than this threshold are discarded. A threshold related to the distribution of the weights. For example, the threshold may be the average weight, a ratio of the average weight, or some percentile of the weights. Features with weights below this threshold are discarded. Discard a set number of features with the lowest weights.

[0119] Because each feature represents a dimension of the model, reducing the number of features through the normalization process reduces the number of dimensions of the model. This results in fewer parameters that need to be sent to the edge device, simplifying the evaluation calculation. This reduces the computational cost, time required, and energy usage to process data on the edge, with minimal impact on segmentation accuracy.

[0120] As part of the logistic regression training process, one of the model parameters learned is a constant bias term, which embodies the overall likelihood of being classified into the seed classification given the classification membership of all known users. This improves accuracy across all similar segments, but especially when there is little or no segment data for a user, the bias term acts as a default probability for predicting similarity based on data for the entire audience.

[0121] In the above embodiment, the users are visitors to an operator's website, but it will be appreciated that the invention is broader in scope and is also applicable to users of, for example, a gaming app or software package.

[0122] The scalability of the disclosed invention makes it highly suitable for IoT devices, allowing for monitoring and reacting to sensor data in near real-time. For example, a data center may have many highly sensitive temperature sensors placed around servers to continuously measure the temperature of electronic chips. Segments can be set for any large deviations in temperature. Reactions may include, for example, increasing fan speed, adjusting the building's air conditioning, or managing processor workload. Average operating temperature and sensor characteristics can be taken into account by machine learning models. When segmentation is performed on the edge, in near real-time, segments can be calculated and reactions can be triggered to ensure that chips are kept at optimal temperatures.

[0123] Another potential use case of the present invention is in anomaly detection, especially when the number of users or devices is very large and anomalies need to be detected quickly. A prime example is in e-commerce, where fraudulent transactions need to be spotted and acted upon as soon as possible. With the advent of smart cards, e-wallets and smartphones with fintech or banking applications, there are many edge devices that could be utilized for this purpose. The integration of machine learning to extend segments in this context means that anomalous transactions can still be found within a segment that would not have been captured from the set segment criteria alone.

[0124] Implementation of the proposed system in the fitness and healthcare device market could be beneficial since personal and sensitive information is stored and evaluated locally. Users may be wary of sharing their personal data with cloud-based systems. The proposed system addresses this privacy concern by ensuring that raw user data (e.g., heart rate, blood pressure) never leaves the personal device and therefore cannot be accessed by third parties such as insurance companies or credit rating agencies.

[0125] 6 illustrates an exemplary electronic device 801 according to any of the electronic devices of the present disclosure (e.g., a third party data provider 1, a machine learning service 2, a cloud online dashboard 4, a client device 6, a classification database 7, an operator device 8, or any one of the devices in the cloud network 3 or the content delivery network 5). The electronic device 801 includes processing circuitry 810 (such as a processor) and memory 812. The electronic device 801 may also include one or more of the following subsystems: a power source 814, a display 816, a wireless transceiver 820, and an input 826.

[0126] The processing circuit 810 can control the operation of the electronic device 801 and the subsystems 812-828. The processor 810 is communicatively coupled to the subsystems 812-828. The memory 812 may include one or more of random access memory (RAM), read-only memory (ROM), non-volatile random access memory (NVRAM), flash memory, other volatile memory, and other non-volatile memory.

[0127] The display 816 may be communicatively coupled to the processing circuit 810, and the processing circuit may be configured to cause the display 816 to output images. The display 816 may include a touch-sensitive interface, such as a touch screen display. The display 816 may be used to interact with software running on the processor 810 of the electronic device 801. The touch-sensitive interface allows a user to provide input to the processing circuit 810 via a discreet touch, multiple touches, or one or more gestures to control the operation of the processing circuit and the functions described herein. It will be appreciated that other forms of input interfaces, such as an input 826, which may include a keyboard or a mouse in the input device, may additionally or alternatively be used for the same purpose.

[0128] The transceiver 820 may include one or more long-range RF transceivers configured to operate according to a communication standard such as LTE, UMTS, 3G, EDGE, GPRS, GSM, and Wi-Fi. For example, the electronic device 801 may include a first long-range radio transceiver 821, such as a cellular transceiver configured to communicate with a cell tower 803 via a cellular data protocol such as LTE, UMTS, 3G, EDGE, GPRS, or GSM, and a second long-range radio transceiver 828, such as a Wi-Fi transceiver configured to communicate with a wireless access point 804 via a Wi-Fi standard such as 802.11ac / n / g / b / a. In this regard, and for purposes of all embodiments herein relating to long-range radio protocols, the long-range radio protocol may be a protocol capable of and designed to communicate over 5 m, 10 m, 20 m, 30 m, 40 m, 50 m, or 100 m. This is in contrast to the short-range radio protocols discussed above. Long-range wireless protocols may utilize higher power for communication than short-range wireless protocols, and the range (e.g., line-of-sight distance) between long-range end nodes (e.g., between an electronic device and a router or base station) of the long-range wireless protocols may be greater than the range (e.g., line-of-sight distance) between short-range end nodes (e.g., between an electronic device and a wireless beacon).

[0129] The electronic device 801 may be configured to communicate with a network 840 via the long-range wireless transceiver 820. The network 840 may be a cloud network 3, or any such wide area network, such as the Internet, or a local area network. The electronic device 801 may be further configured to communicate with one or more servers or user devices via the long-range wireless transceiver 820 and the network 840. These servers or user devices may be any one of those described herein.

[0130] The term "comprising" encompasses "including" and "consisting", e.g., a composition "comprising" X can consist only of X, or it can include something additional, e.g., X+Y.

[0131] Unless otherwise indicated, each embodiment described herein can be combined with other embodiments as described herein.

[0132] The methods described herein may be implemented by software in machine-readable form on a tangible storage medium, e.g., in the form of a computer program, the computer program including computer program code means adapted to perform all the steps of any of the methods described herein when the program is executed on a computer and the computer program may be embodied on a computer readable medium. Examples of tangible (or non-transitory) storage media include disks, thumb drives, memory cards, etc., and do not include propagated signals. The software may be suitable for execution on parallel or serial processors such that the method steps may be performed in any suitable order or simultaneously. This acknowledges that firmware and software may be valuable, separately tradable goods. It is intended to encompass software that operates on or controls "dumb" or standard hardware to perform a desired function. It is also intended to encompass software that "describes" or defines the configuration of hardware, such as HDL (hardware description language) software, such as used to design silicon chips or configure universal programmable chips, to perform a desired function.

[0133] It will be understood that the modules described herein may be implemented in hardware or software. Further, the modules may be implemented in various locations throughout the system.

[0134] Those skilled in the art will understand that storage devices used to store program instructions can be distributed across a network. For example, a remote computer can store an example of a process described as software. A local or terminal computer can access a remote computer and download some or all of the software to execute the program. Alternatively, a local computer can download parts of the software as needed, or execute some software instructions at a local terminal and some software instructions at a remote computer (or computer network). Those skilled in the art will also understand that by utilizing conventional techniques known to those skilled in the art, all or part of the software instructions can be executed by dedicated circuitry such as DSPs, programmable logic arrays, etc.

[0135] As will be apparent to one skilled in the art, any ranges or device values ​​given herein can be expanded or modified without losing the effect sought.

[0136] It will be understood that the benefits and advantages described above may relate to one embodiment or to several embodiments, and the embodiments are not limited to those that solve any or all of the problems described or those that have any or all of the benefits and advantages described.

[0137] A reference to "an" item refers to one or more of those items. The term "comprising" is used herein to mean including specified method blocks or elements, but such blocks or elements do not constitute an exclusive list and the method or apparatus may include additional blocks or elements.

[0138] The steps of the methods described herein may be performed in any suitable order, or simultaneously where appropriate. Additionally, individual blocks may be deleted from any method without departing from the spirit and scope of the subject matter described herein. Aspects of any of the above embodiments may be combined with aspects of any of the other embodiments described above to form further embodiments without losing the desired effect. Any of the above modules may be implemented in hardware or software.

[0139] It will be understood that the above description of preferred embodiments is given by way of example only, and that various modifications may be made by those skilled in the art. 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 may make numerous changes to the disclosed embodiments without departing from the scope of the invention.

[0140] [Embodiment] (1) In a system for classifying an entity or classifying the entity's similarity to other entities, a server, the server comprising: a storage means configured to store characteristics of a first plurality of entities, each of said first plurality of entities being classified in a first classification; a training module configured to train a classification model to classify an input entity with the first classification or to classify a similarity between the input entity and the first plurality of entities based on characteristics of the input entity, the classification model being trained using characteristics of the first plurality of entities as training data; a transmitter configured to transmit the trained classification model to a client device; Including, The system further includes a client device, the client device comprising: a receiver configured to receive the trained classification model; a memory configured to store the trained classification model; and a processor configured to process data comprising a characteristic of at least one input entity using the trained classification model stored on the client device and cause the trained classification model to output a signal, the signal comprising: classifying said at least one input entity with said first classification, or and a processor that classifies an affinity between the at least one input entity and the first plurality of entities. (2) The system of embodiment 1, wherein the trained classification model is configured to output a similarity score to classify the similarity between an input entity and the first plurality of entities, and optionally, a receiver in the client device is configured to receive a threshold similarity score, and the processor is configured to perform an action if the similarity score of the at least one input entity satisfies the threshold similarity score. (3) The system of embodiment 2, wherein the processor is configured to perform a plurality of different actions at the client device, each of the actions being performed in response to a different similarity score being associated with the at least one input entity, and / or the receiver at the client device is configured to receive a threshold similarity score, and the processor is configured to classify the at least one input entity with the first classification if the similarity score of the at least one input entity satisfies the threshold similarity score. (4) the receiver is configured to obtain, at the server, characteristics of a second plurality of entities that are different from the first plurality of entities; the server further comprises a model analyzer configured to process data including the characteristics of the second plurality of entities using the trained classification model to output a similarity score for each of the second plurality of entities; A system as described in any of embodiments 1 to 3, wherein each similarity score output by the trained classification model is associated with one of the second plurality of entities and classifies the similarity between the associated entity and the first plurality of entities, and optionally the model analyzer is configured to calculate the number of the second entities associated with each similarity score output by the trained classification model, and optionally the receiver in the client device is configured to receive a threshold similarity score set based on the number of the second entities associated with each similarity score. (5) The system further includes an operator device, the operator device comprising: an interface configured to allow an operator to specify a first set of criteria for comparison with characteristics of entities, entities that satisfy the first set of criteria being classified with the first classification; and a transmitter configured to transmit the first set of criteria to the server; Including, The server further includes a receiver configured to receive the first set of criteria; A system as described in any of embodiments 1 to 4, wherein the training module is configured to compare the first set of criteria with characteristics of a plurality of entities and classify entities that satisfy the first set of criteria with the first classification, and optionally the training module is configured to calculate a number of entities that satisfy the first set of criteria, and the training module is configured to start training the classification model based on the number of entities that satisfy the first set of criteria.

[0141] (6) A system as described in any one of embodiments 1 to 5, wherein each of the entities is a computing device and the characteristics respectively describe the performance of each computing device. (7) A system described in any one of embodiments 1 to 6, wherein each entity is a user and the characteristics respectively describe attributes of each user. (8) The system of any of embodiments 1 to 7, wherein training the classification model includes determining a weighting for each of a plurality of parameters, the weighting for each of the parameters indicating the magnitude of influence the parameter has on the output signal, and optionally, the processor in the training module or the client device is configured to determine that at least one parameter is associated with a weighting that does not meet a threshold weighting, and optionally, the processor is configured to prevent the trained classifier from using the at least one parameter associated with a weighting that does not meet a threshold weighting. (9) The system of embodiment 8, wherein the training module is configured to omit from the trained model at least one parameter associated with a weighting that does not meet the threshold weighting, and / or the training module is configured to omit from the trained model at least one parameter associated with a weighting that does not meet the threshold weighting by not sending the at least one parameter associated with a weighting that does not meet the threshold weighting to the client device. (10) the training module is configured to retrain the classification model to classify an input entity with the first classification or classify a similarity between the input entity and the first plurality of entities, the classification model being retrained using a second different set of characteristics of the plurality of entities classified with the first classification as training data, optionally the classification model being retrained using a second different set of characteristics of a plurality of entities classified with the first classification that are different from the first plurality of entities, and / or 10. The system of any one of embodiments 1 to 9, wherein the training module is configured to retrain the classification model at a predetermined frequency.

[0142] (11) The transmitter is configured to transmit the retrained classification model to the client device; the receiver at the client device is configured to receive the retrained classification model; the memory is configured to store the retrained classification model; The processor is configured to use the retrained classification model stored on the client device to process data comprising characteristics of at least one input entity to cause the retrained classification model to output a signal, the signal comprising: classifying said at least one input entity with said first classification, or 11. The system of embodiment 10, further comprising classifying a similarity between the at least one input entity and the first plurality of entities. (12) The system of embodiment 11, wherein the transmitter is configured to transmit the retrained classification model to the client device by only transmitting parameters of the retrained model that differ from parameters of the trained model. (13) The processor in the client device is configured to determine whether an input entity is classified with an initial classification; 13. The system of any of claims 1 to 12, wherein the processor is configured to process data including at least one characteristic of the input entity using the stored classification model only if the stored classification model differs from the classification model used to classify the input entity in the initial classification. (14) The system of any one of embodiments 1 to 13, wherein the processor in the client device is configured to determine whether an input entity is classified with an initial classification, obtain a retrained classification model from the server only if a most recently trained classification model at the server differs from the classification model used to classify the input entity with the initial classification, and use the retrained classification model to process data including at least one characteristic of the input entity. (15) A computer-implemented method for generating a trained classification model for classifying an entity or classifying the entity's similarity to other entities, comprising: obtaining, at a server, a characteristic of a first plurality of entities, each of the first plurality of entities being classified with a first taxonomy; training a classification model at the server to classify an input entity with the first classification or classify a similarity between the input entity and the first plurality of entities based on characteristics of the input entity, the classification model being trained using characteristics of the first plurality of entities as training data; transmitting the trained classification model from the server to the client device for use at the client device; A method comprising:

Claims

1. 1. A system for classifying an entity or classifying its similarity to other entities, comprising: a server, the server comprising: a storage means configured to store characteristics of a first plurality of entities transmitted from a client device, each of the first plurality of entities being classified according to a first classification; a training module configured to train a classification model to classify an input entity with the first classification or to classify a similarity between the input entity and the first plurality of entities based on characteristics of the input entity, the classification model being trained using characteristics of the first plurality of entities as training data; a transmitter configured to transmit the trained classification model to a client device; Including, The system further includes a client device, the client device comprising: a receiver configured to receive the trained classification model; a memory configured to store the trained classification model; and a processor configured to process data comprising characteristics of at least one input entity using the trained classification model stored on the client device and cause the trained classification model to output a signal, the signal comprising: classifying said at least one input entity with said first classification, or and a processor for classifying an affinity between the at least one input entity and the first plurality of entities.

2. 2. The system of claim 1, wherein the trained classification model is configured to output a similarity score to classify a similarity between an input entity and the first plurality of entities, and optionally, a receiver in the client device is configured to receive a threshold similarity score, and the processor is configured to perform an action if the similarity score of the at least one input entity satisfies the threshold similarity score.

3. 3. The system of claim 2, wherein the processor is configured to perform a plurality of different actions at the client device, each of the actions being performed in response to a different similarity score being associated with the at least one input entity, and / or the receiver at the client device is configured to receive a threshold similarity score, and the processor is configured to classify the at least one input entity with the first classification if the similarity score of the at least one input entity satisfies the threshold similarity score.

4. The receiver is configured to obtain, at the server, a characteristic of a second plurality of entities that is different from the first plurality of entities; the server further comprises a model analyzer configured to process data including the characteristics of the second plurality of entities using the trained classification model to output a similarity score for each of the second plurality of entities; 4. The system of claim 1, wherein each similarity score output by the trained classification model is associated with one of the second plurality of entities and classifies a similarity between the associated entity and the first plurality of entities, and optionally the model analyzer is configured to calculate a number of the second entities associated with each similarity score output by the trained classification model, and optionally the receiver in the client device is configured to receive a threshold similarity score set based on the number of the second entities associated with each similarity score.

5. The system further includes an operator device, the operator device comprising: an interface configured to allow an operator to specify a first set of criteria for comparison with characteristics of entities, entities that satisfy the first set of criteria being classified with the first classification; and a transmitter configured to transmit the first set of criteria to the server; Including, The server further includes a receiver configured to receive the first set of criteria; 5. The system of claim 1, wherein the training module is configured to compare the first set of criteria with characteristics of a plurality of entities and classify entities that satisfy the first set of criteria with the first classification, and optionally the training module is configured to calculate a number of entities that satisfy the first set of criteria, and the training module is configured to start training the classification model based on the number of entities that satisfy the first set of criteria.

6. The system of claim 1 , wherein each said entity is a computing device, and each said characteristic describes a performance of each respective computing device.

7. The system of claim 1 , wherein each entity is a user, and the characteristics describe attributes of each user, respectively.

8. 8. The system of claim 1, wherein training the classification model comprises determining a weighting for each of a plurality of parameters, the weighting for each of the parameters indicating the magnitude of influence the parameter has on the output signal, and optionally the processor in the training module or the client device is configured to determine that at least one parameter is associated with a weighting that does not meet a threshold weighting, and optionally the processor is configured to prevent the trained classifier from using the at least one parameter associated with a weighting that does not meet a threshold weighting.

9. 10. The system of claim 8, wherein the training module is configured to omit from the trained classification model the at least one parameter associated with a weighting that does not meet the threshold weighting, and / or the training module is configured to omit from the trained classification model the at least one parameter associated with a weighting that does not meet the threshold weighting by not sending the at least one parameter associated with a weighting that does not meet the threshold weighting to the client device.

10. the training module is configured to retrain the classification model to classify an input entity with the first classification or to classify a similarity between the input entity and the first plurality of entities, the classification model being retrained using as training data a second different set of characteristics of the plurality of entities classified with the first classification, optionally the classification model being retrained using as training data a second different set of characteristics of a plurality of entities classified with the first classification that are different from the first plurality of entities, and / or The system of claim 1 , wherein the training module is configured to retrain the classification model at a predetermined frequency.

11. the transmitter is configured to transmit the retrained classification model to the client device; the receiver at the client device is configured to receive the retrained classification model; the memory is configured to store the retrained classification model; The processor is configured to use the retrained classification model stored on the client device to process data comprising characteristics of at least one input entity to cause the retrained classification model to output a signal, the signal comprising: classifying said at least one input entity with said first classification, or The system of claim 10 , further comprising: classifying an affinity between the at least one input entity and the first plurality of entities.

12. 12. The system of claim 11, wherein the transmitter is configured to transmit the retrained classification model to the client device by only transmitting parameters of the retrained classification model that differ from parameters of the trained classification model.

13. The processor at the client device is configured to determine whether an input entity is classified with an initial classification; 13. The system of claim 1, wherein the processor is configured to use the stored classification model to process data comprising at least one characteristic of an input entity only if the stored classification model differs from the classification model used to classify the input entity in the initial classification.

14. 14. The system of claim 1, wherein the processor at the client device is configured to determine whether an input entity has been classified with an initial classification, and to obtain a retrained classification model from the server only if a most recently trained classification model at the server differs from the classification model used to classify the input entity with the initial classification, and to process data including at least one characteristic of the input entity using the retrained classification model.

15. 1. A computer-implemented method for generating a trained classification model for classifying an entity or classifying similarity of said entity to other entities, comprising: At the server, acquiring characteristics of a first plurality of entities transmitted from the client device, each of the first plurality of entities being classified with a first classification; training a classification model at the server to classify an input entity with the first classification or classify a similarity between the input entity and the first plurality of entities based on characteristics of the input entity, the classification model being trained using characteristics of the first plurality of entities as training data; transmitting the trained classification model from the server to the client device for use at the client device; A method comprising:

Images