Abstract: Devices and techniques are generally described for fraud detection. A machine learning model is used to determine a first fraud risk score for a first transaction. The machine learning model includes a first set of weights. A first covariance matrix is determined for the machine learning model based at least in part on the first fraud risk score. A second set of weights for the machine learning model is determined. The second set of weights is determined based on the first set of weights and the first covariance matrix. In various examples, the machine learning model with the second set of weights is used to determine a second fraud risk score for a second transaction. A fraud decision surface is determined and the second fraud risk score is compared to the fraud decision surface. Data indicating that the second transaction is fraudulent is sent to a computing device.

Abstract: One embodiment provides a system for facilitating anomaly detection. During operation, the system determines, by a computing device, a set of training instances, wherein a training instance represents a single class of data within a predefined range. The system computes a similarity score for each testing instance in a set of testing instances, wherein the similarity score is based on a similarity function which takes as input a respective testing instance and the set of training instances. The system determines a boundary threshold based on an ordering of the similarity score for each testing instance. The system classifies a first testing instance as an anomaly responsive to determining that the first testing instance falls outside the boundary threshold, thereby enhancing data mining and outlier detection in the single class of data using unlabeled training instances.

Abstract: A device that includes an enterprise data indexing engine (EDIE) configured to receive a set of sentences and to compare the words in the sentences to a set of predefined keywords. The EDIE is further configured to identify one or more sentences that do not contain any of the keywords and to associate the identified sentences with a first classification type. The EDIE is further configured to identify a sentence that contains one or more keywords and to associate the sentence with a second classification type. The EDIE is further configured to link together the sentence that is associated with the second classification type and the sentences that are associated with the first classification type.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for rule generation and interaction. A rules engine is provided that does not require manual effort to generate or maintain high-quality rules over time for detecting malicious accounts or events. Rules no longer need to be manually added, tuned, or removed from the system. The system is able to determine the health of each rule and automatically add, tune, and remove rules to maintain a consistent, effective rule set.

Abstract: A non-transitory computer-readable recording medium storing a program that causes a computer to execute a process for evaluating a voice, the process includes analyzing a voice signal to detect a pitch frequency; selecting an evaluation target region to be evaluated in the detected pitch frequency based on a distribution of a detection rate of the detected pitch frequency; and evaluating a voice based on the distribution of the detection rate and the selected evaluation target region. An Impression of the voice using a correction to the distribution is evaluated and determined as good when a spread of the corrected distribution is larger than or equal to a certain threshold.

Abstract: This disclosure relates to methods, non-transitory computer readable media, and systems that, upon identifying a set of distribution contacts, generate clusters of distribution contacts from a sampled subset of distribution contacts and assign remaining distribution contacts from the set to the generated clusters for visualization in a user interface. By clustering a representative sample of such distribution contacts, the disclosed methods, non-transitory computer readable media, and systems can quickly analyze and identify contact characteristics in clusters of distribution contacts, including common contact characteristics exhibited by a given cluster's contacts. The disclosed methods, non-transitory computer readable media, and systems can accordingly respond to user requests for a cluster analysis by expeditiously generating cluster visualizations identifying contact characteristics of clustered distribution contacts.

Abstract: Media items associated with status values are stored using a server in a protected environment. A device outside the protected environment requests sending media items to a client device that is also outside. Using machine learning systems that can be trained using attribute values associated with personal data records to output the existence of a status value, the server computer trains a particular machine learning system in the protected environment only if specified data meets specified criteria, then sends the resulting trained ML system to the requesting device. That device evaluates the trained ML system to determine which media items to deliver to the client device in the manner set forth in the claims.

Abstract: Within a portion of a currently executing compiled streams application according to a request to modify a configuration of the portion, a placement is performed of an operator of the currently executing compiled streams application into a processing element, wherein the currently executing compiled streams application has a configuration assigned before application execution began. In response to the placement, a resource allocation corresponding to the placement is performed according to the request, the resource allocation specifying a resource assigned to the processing element. The processing element is activated to execute the operator according to the placement and the resource allocation.

Abstract: An approach is provided that detects transmission of sensitive data fields from a user device to an online site. The approach determines the data types of the sensitive data fields and gathers site information from the online site which are stored in a data store along with the current date. Subsequently, the user can manage the data by displaying site information on a display of the user device. Each record of site information pertains to one of many online sites including the selected online site. The user makes a data management request at the user device pertaining to the selected site information. The approach then responsively transmits a request to the selected online site with the request being based on the received data management request. One example of a request is a deletion request that requests that the online site remove the user's sensitive data from the online site.

Abstract: Techniques to correlate event data are disclosed. In various embodiments, an aggregation engine is used to correlate event data received from one or more source systems based on one or more correlation rules. An event group comprising at least a portion of said correlated event data is generated programmatically and is asserted as a fact in a working memory of a Rete engine configured to apply one or more Rete rules to facts in the working memory.

Abstract: A method, system and computer program product for generating an optimization model. A prescriptive model including a structured set of data and instance(s) of templates of a technical and/or business prescriptive domain is received, where the instances of the templates define goals and/or constraints for the structured set of data. The business prescriptive domain extends the technical prescriptive domain defining a generic optimization problem and the business prescriptive domain defines a more specific business problem. The prescriptive model is translated into a technical prescriptive model using a set of mapping rules to translate the instance(s) of the templates of the business prescriptive domain into instance(s) of the templates of the technical prescriptive domain from which the business prescriptive domain extends. Furthermore, the technical prescriptive model is translated into an optimization model, where the optimization model is solved.

Abstract: An image classification system includes a storage device, a computing device and a first processing device. The storage device stores a plurality of pseudo-centroid datasets, wherein the pseudo-centroid datasets correspond to a plurality of units of first image dataset, and the number of pseudo-centroid data points of each of the pseudo-centroid datasets is much smaller than the number of data points of each of the units of first image dataset. The computing device receives the second image data and computes a plurality of feature values of the second image data. The first processing device receives the feature values and the pseudo-centroid datasets, and compares the feature values with the pseudo-centroid data points to identify and classify the second image data.

Abstract: The present disclosure provides an information extraction method. The method includes obtaining an unstructured text data; parsing the unstructured text data based on a node format; and generating a first node set that comprises nodes describing the unstructured text data. The method further includes obtaining a rule database, the rule database comprising a plurality of rules used for generating a node, each rule designating a role of a node generated by using the rule, and the role of the node being a root node or a non-root node; synthesizing the nodes in the first node set based on the rule database to generate a root node; and generating structured data based on the root node.

Abstract: Methods, systems, and computer program products are provided. A network device receives, from a client device, a description of a dataset to be de-identified, and a list of one or more data de-identification techniques selected from groups consisting of a group of data masking techniques and a group of data pseudonymization techniques, and their configuration options supported by the client device. A first technique, from the at least one group of techniques and its configuration options supported by the client device and the network device are determined. The network device receives a dataset produced at the client device by applying the first technique and selected configuration options to corresponding attributes from the client device. The network device applies a de-identification technique to the dataset to produce a resulting set of de-identified data, wherein the de-identification technique is coordinated with the first technique and its configuration options to de-identify the dataset.

Abstract: Embodiments for localizing abnormal energy consumption at a facility in a cloud computing environment by a processor. One or more residuals for both one or more predictors and a prediction may be generated according to one or more energy consumption measurements, weather data, and one or more characteristics of the one or more facilities, or a combination thereof. An energy consumption anomaly may be localized according to those of the one or more residuals associated with one or more predictors having an actual energy measurement deviating from a predicted actual energy measurement.

Abstract: The present invention relates to a method for composing an application using a plurality of distributed interaction patterns including services, events, actions and properties. A second semantic model or Thing Recipe is provided as a basis for representing the application. This Thing Recipe provides a concept for orchestration or choreography of interaction patterns distributed across a network. The process from a specification of the Thing Recipe to its deployment is considerable shorter than the implementation and orchestration of web service. The Thing Recipe is not only interpretable but also executable. The semantic approach of the Thing Recipe enables a formal representation capable of a machine-based discovery of the application functionality and, at the same time, a machine interpretation and execution. This means that a Thing Recipe can be directly executed in any device that has an embedded runtime with reasoning capabilities.

Abstract: Certain embodiments involve determining and outputting correlations between metrics in large-scale web analytics datasets. For example, a processor identifies pairs of data metrics in a web analytics data set and determines a Maximal Information Coefficient (MIC) score for each pair of data metrics that indicates a strength of a correlation between the pair of data metrics. The processor generates an interactive user interface that graphically displays each pair of correlated data metrics having an MIC score above a threshold and the interactive user interface indicates the strength of the correlation between each displayed pair of correlated data metrics. The processor receives user input indicating an adjustment to the threshold and modifies the interactive user interface in response to receiving the user input by adding pairs of correlated data metrics to, or removing pairs of correlated metrics from, the interactive user interface based on the adjustment to the threshold.

Abstract: Patient treatment may be optimized using Recurrent Neural Network (RNN) based state simulation and Reinforcement learning (RL) techniques to simulate future states and actions. A RNN state simulator and a RL action generator may be trained using patient data such as historical states and actions. The RL action generator may be optimized by applying the RNN state simulator to simulating future states and applying the RL action generator to generate recommended actions based on the simulated future states. This process may be iteratively performed until a computational convergence is reached by the RL action generator which may indicate that the RL action generator has been optimized. A patient state may be fed into the optimized RL action generator to generate an optimal recommended treatment action.

Abstract: In one embodiment, a method includes accessing policy states associated with a policy, determining that a particular policy state of the policy states should not be pruned, and generating possible next policy states for the particular policy state based on one or more policy actions and one or more costs associated with the respective policy actions, wherein each policy action is associated with one or more probabilities corresponding to one or more possibilities of resulting in one or more of the possible next policy states, wherein the respective cost associated with the policy action is determined based on one or more computing costs of the policy action resulting in the one or more of the possible next policy states in combination with the respective one or more probabilities, and wherein each policy state of the created policy evaluation tree comprises a rule state for each rule of the policy.

Abstract: An example method for assisting a patient to conduct a physical therapy session can include: presenting information about the physical therapy session on a display, the information including one or more actions to be performed by the patient; capturing one or more images of the patient as the patient performs the one or more actions; detecting, by a computing device, a safety risk related to one or more medical devices associated with the patient during the physical therapy session; and providing feedback regarding the safety risk. Other methods can include: detecting, by a computing device, using the one or more images, a level of effort from the patient during the physical therapy session; and modifying a future physical therapy session for the patient based upon the level of effort; or allowing the patient to compete with other patients during the physical therapy session based upon the level of effort.

Abstract: An example integrated circuit includes a memory including a non-deterministic finite automata (NFA) buffer configured to store a plurality of instructions defining an ordered sequence of instructions of at least a portion of an NFA graph, the portion of the NFA graph comprising a plurality of nodes arranged along a plurality of paths. The NFA engine determines a current symbol and one or more subsequent symbols of a payload segment that satisfy a match condition specified by a subset of instructions of the plurality of instructions for a path of the plurality of paths and in response to determining the current symbol and the one or more subsequent symbols of the payload segment that satisfy the match condition, outputs an indication that the payload data has resulted in a match.

Abstract: A method for improving memory efficiency of production rule systems is described. In one embodiment, the method includes identifying a rule associated with production rule systems, constructing a production rule network based at least in part on the rule, identifying a positional constraint associated with the rule, and implementing an alpha memory gate in the production rule network based at least in part on the positional constraint. In some cases, the alpha memory gate is one of a plurality of nodes of the production rule network.

Abstract: The present invention relates to a method and system for determining the quality of an application based on user behaviors of application management. The method includes: 1) preprocessing data on user behaviors of application management to obtain a sequence of management behaviors as a training set; 2) extracting features of the user behaviors from the obtained sequence of management behaviors; 3) establishing a multi-dimensional independent variable space using the extracted behavior features, wherein each application in the training set is considered as a point in the space; training a model according to the space; and debugging to obtain optimal model parameters and a definite regression model; 4) arranging the data of the management behaviors of each application to be predicted according to step 1) and step 2), and then inputting the result into the regression model, wherein an output of the regression model is the result of the application quality determination.

Abstract: Methods and systems for a quantum computing approach to solving challenging, e.g., NP-complete, problems in transportation. One of the methods includes (a) ingesting transportation-related data into a graph structure, the transportation-related data being associated with a transportation system; (b) identifying a transportation metric associated with the transportation system; (c) identifying at least one attribute associated with the transportation-related data, where the transportation metric is based at least in part on the attribute; (d) using a quantum computer to derive an operational parameter for the attribute that improves the transportation metric; and (e) applying the operational parameter to the operation of the transportation system.

Abstract: Exemplary methods, apparatuses, and systems receive, from a first user device associated with a first user, location data representing a current physical location of the first user. A request for the current physical location of the first user is received from a second user device associated with the second user. In response to the received selection, the current physical location of the first user based upon the shared location data is transmitted to the second user device. A selection of a third party service is received from the second user device. The current physical location of the first user is transmitted to a server associated with the third party service within a request for the third party service to send the a service or product to the current physical location of the first user.

Abstract: A method of unlocking an electronic device, an unlocking device and system and a storage medium are provided. The method includes: acquiring at least one image to be authenticated, in which the image to be authenticated is an RGB image or an infrared image; obtaining an RGB image authentication result of the image to be authenticated by RGB object authentication of the image to be authenticated in a case where the image to be authenticated is the RGB image, and obtaining an infrared image authentication result of the image to be authenticated by infrared object authentication of the image to be authenticated in a case where the image to be authenticated is the infrared image; and determining whether to unlock the electronic device according to at least one of the RGB image authentication result and the infrared image authentication result.

Abstract: The disclosure provides an approach for learning to schedule control fragments for physics-based virtual character simulations and physical robot control. Given precomputed tracking controllers, a simulation application segments the controllers into control fragments and learns a scheduler that selects control fragments at runtime to accomplish a task. In one embodiment, each scheduler may be modeled with a Q-network that maps a high-level representation of the state of the simulation to a control fragment for execution. In such a case, the deep Q-learning algorithm applied to learn the Q-network schedulers may be adapted to use a reward function that prefers the original controller sequence and an exploration strategy that gives more chance to in-sequence control fragments than to out-of-sequence control fragments.

Abstract: Systems, devices, media and methods are presented for splitting decision trees between server and client. The client of the systems and methods sends a configuration query. The server of the system and method receives the configuration query. The server retrieves Config rule(s) according to the configuration query. Each of the Config rule(s) can be represented by decision tree(s). The server evaluates the decision tree(s). If a definitive True or False cannot be derived from the evaluation using server knowledge, the server prunes the decision tree(s) and returns them to client side for further evaluation.

Abstract: Embodiments for detection of energy consumption anomalies in one or more energy consumption systems in a cloud computing environment by a processor. Energy consumption may be predicted for one or more facilities according to one or more energy consumption measurements, weather data, and one or more characteristics of the one or more facilities, or a combination thereof. An onset of an energy consumption anomaly may be detected according to the prediction.

Abstract: Example embodiments relate to parallelizing structured query language (SQL) user defined transformation functions. In example embodiments, a subquery of a query is received from a query engine, where each of the subqueries is associated with a distinct magic number in a magic table. A user defined transformation function that includes local, role-based functionality may then be executed, where the magic number triggers parallel execution of the user defined transformation function. At this stage, the results of the user defined transformation function are sent to the query engine, where the query engine unions the results with other results that are obtained from the other database nodes.

Abstract: Techniques are disclosed for standardization of data. According to a first technique, standard representation terms are determined for to-be-standardized data using the to-be-standardized data itself and without using any external reference data. According to a second technique, a combination of the to-be-standardized data and an external reference is used to determine standard representation terms for the to-be-standardized data.

Abstract: Patient treatment may be optimized using Recurrent Neural Network (RNN) based state simulation and Reinforcement learning (RL) techniques to simulate future states and actions. A RNN state simulator and a RL action generator may be trained using patient data such as historical states and actions. The RL action generator may be optimized by applying the RNN state simulator to simulating future states and applying the RL action generator to generate recommended actions based on the simulated future states. This process may be iteratively performed until a computational convergence is reached by the RL action generator which may indicate that the RL action generator has been optimized. A patient state may be fed into the optimized RL action generator to generate an optimal recommended treatment action.

Abstract: An illicit website identifying (IWI) computing device is described herein, the IWI computing device configured to store a log of suspect illicit websites, the log including website content associated with each suspect illicit website. Each suspect illicit website is associated with a respective merchant. The IWI computing device is also configured to receive transaction data for at least one transaction initiated at a first suspect illicit website, the first suspect illicit website included within the log of suspect illicit websites, and analyze the received transaction data to identify one or more transactions including at least a portion of the website content associated with the first suspect illicit website. The IWI computing device is further configured to identify the first suspect illicit website and the merchant associated therewith as illicit for further investigation, and generate an identification report including the identified illicit website and merchant.

Abstract: Aspects of the disclosed technology provide a method including receiving one or more signals emanated from a monitored device; signal processing, based on a software model and a hardware-software (HW/SW) interaction model of the monitored device, the one or more signals to determine if an anomaly exists in the one or more signals; and responsive to determining, based on the signal processing, that an anomaly exists, transmitting an indication of the anomaly.

Abstract: In accordance with embodiments, there are provided mechanisms and methods for managing access to data based on information associated with a physical location of a user. These mechanisms and methods for managing access to systems, products, or data based on information associated with a physical location of a user can enable improved data management efficiency, enhanced data management accuracy, decreased data management costs, decreased licensing costs, increased security, additional marketing opportunities, etc.

Abstract: Provided is an. emotion. estimating apparatus capable of precisely estimating an emotion and mental state of a measurement subject by using a non-contact pulse detection technology. The emotion estimating apparatus forcibly performs re-sampling processing of digital biometric data generated by converting a heart rate signal to digital data after extracting data for one cycle at an RR interval, and obtains coefficients of harmonic components by a DCT conversion processing unit. From the coefficients of the harmonic components, AC components are removed by LPFs, whereby a coefficient data array is obtained.. The coefficient data array is compared to a dictionary data group, in which dictionary data is a characteristic quantity indicating an emotion or mental state, and a similarity therebetween is calculated, whereby an emotion or mental state of a subject is estimated on the basis of the heart rate signal of the subject.

Abstract: According to examples, an apparatus may include a machine readable medium on which is stored machine readable instructions that may cause a processor to, for each of a plurality of resource setting levels, determine resource usage characteristics and execution times of executed workloads, assign, based on the resource usage characteristics of the executed workloads, each of the executed workloads into one of a plurality of resource bins, determine, for each of the resource bins, an average execution time of the executed workloads in the resource bin, determine a total average execution time of the executed workloads from the determined average execution times, identify a lowest total average execution time of the determined total average execution times, determine the resource setting level corresponding to the identified lowest total average execution time, and tune a resource setting to the determined resource setting level.

Abstract: The invention relates to a method for triggering one or several actions complementary to those triggered by an application method implemented by an electronic device. The invention further relates to a system including one or several electronic devices implementing said method triggering complementary actions to cooperate with one or several communicating electronic peripherals.

Abstract: Methods, systems, and computer-readable storage media for ranking anomaly detection algorithms, including operations of receiving a set of unlabeled data from one or more sensors in a plurality of sensors of an internet of things, generating a plurality of data distributions corresponding to the set of unlabeled data by using a plurality of anomaly detection algorithms, and ranking the plurality of anomaly detection algorithms relative to the set of unlabeled data based on a distance between a first quantile and a second quantile of each of the plurality of data distributions.

Abstract: A first instruction is sent to an integrated smart meter to initialize the integrated smart meter. A second instruction is sent to the integrated smart meter to establish a connection with an energy consuming device. The established connection between the integrated smart meter and the energy consuming device is determined to be acceptable. A third instruction is sent to the integrated smart meter to monitor the energy consuming device. Energy consumption data of the energy consuming device is received from the integrated smart meter.

Abstract: A cognitive information processing system environment comprising: a plurality of data sources, at least some of the plurality of data sources comprising healthcare relevant data sources; a cognitive inference and learning system coupled to receive a data from the plurality of data sources, the cognitive inference and learning system processing the data from the plurality of data sources to provide cognitively processed healthcare relevant insights, the cognitive inference and learning system further comprising performing a learning operation to iteratively improve the cognitively processed healthcare relevant insights over time; and, a destination, the destination receiving the cognitively processed healthcare relevant insights.

Abstract: A system and method for method for generating a security rule classification model comprises receiving at least one security rule from at least one attack database of a first security product of a plurality of different security products; normalizing each of the at least one security rule; generating a vector for each of the least one normalized security rule; classifying each generated vector to a security engine within a security service using a classification sub-model to generate a preliminary classification model, wherein the classification sub-model is provided from previous classification of security rules for a security product of the plurality of different security products that is different than the first security product; determining a score for the preliminary classification model; and validating the preliminary classification model as the security rule classification model, when the score is over a predefined threshold.

Abstract: An aspect of the invention includes reading a scale in image data representing an image of physical characteristics and resizing at least a portion of the image data to align with target image data representing a target image based at least in part on the scale to form resized image data representing one or more resized images. Noise reduction is applied to the resized image data to produce test image data representing one or more test images. A best fit analysis is performed on the test image data with respect to the target image data. Test image data having the best fit are stored with training image data representing classification training images indicative of one or more recognized features. An anomaly in unclassified image data representing an unclassified image is identified based at least in part on an anomaly detector as trained using the classification training images.

Abstract: An example processing device includes a memory including a non-deterministic finite automata (NFA) buffer configured to store a plurality of instructions defining an ordered sequence of instructions of at least a portion of an NFA graph, the portion of the NFA graph comprising a plurality of nodes arranged along a plurality of paths. The NFA engine determines a current symbol and one or more subsequent symbols of a payload segment that satisfy a match condition specified by a subset of instructions of the plurality of instructions for a path of the plurality of paths and in response to determining the current symbol and the one or more subsequent symbols of the payload segment that satisfy the match condition, outputs an indication that the payload data has resulted in a match.

Abstract: An architecture and processes enable computer learning and developing an understanding of arbitrary natural language text through collaboration with humans in the context of joint problem solving. The architecture ingests the text and then syntactically and semantically processes the text to infer an initial understanding of the text. The initial understanding is captured in a story model of semantic and frame structures. The story model is then tested through computer generated questions that are posed to humans through interactive dialog sessions. The knowledge gleaned from the humans is used to update the story model as well as the computing system's current world model of understanding. The process is repeated for multiple stories over time, enabling the computing system to grow in knowledge and thereby understand stories of increasingly higher reading comprehension levels.

Abstract: According to an embodiment, a storage controller comprises a circuitry configured to implement an address generator, a reader, and a duplication detector. The address generator is configured to generate a scan address indicating each storage area of a storage that stores therein externally written data, according to a particular scan pattern for defining an order of an address of data to be read. The reader is configured to read data from the storage area of the storage indicated by the scan address. The duplication detector is configured to detect whether the data read by the reader is a duplicate of any one of a past predetermined number of pieces of data.

Abstract: Computer implemented methods for identifying denial constraints are provided herein. The denial constraints can be used with a database schema R. A predicate space P can be generated for an instance I in the schema R. An evidence set EviI can be generated. The evidence set EviI can include sets of satisfied predicates in the predicate space P for each instance I. A minimal set of predicates can be identified for the evidence set EviI. Valid denial constraints can be identified from the minimal set by inverting the predicates in the minimal set.

Abstract: Penetration testing campaigns generate remediation recommendations based at least in part on information about files stored in network nodes of the tested networked system. Information is obtained about files stored in a plurality of network nodes of the networked system, and based on the obtained information, a corresponding data-value score for each network node of the plurality of network nodes is determined according to a common data-value metric. The penetration testing campaign is executed, following which one or more remediation recommendations are selected based on the data-value scores corresponding to at least some of the plurality of network nodes.

Abstract: The present disclosure relates a system, method, and computer program for detecting anomalous user network activity based on multiple data sources. The system extracts user event data for n days from multiple data sources to create a baseline behavior model that reflects the user's daily volume and type of IT events. In creating the model, the system addresses data heterogeneity in multi-source logs by categorizing raw events into meta events. Thus, baseline behavior model captures the user's daily meta-event pattern and volume of IT meta events over n days. The model is created using a dimension reduction technique. The system detects any anomalous pattern and volume changes in a user's IT behavior on day n by comparing user meta-event activity on day n to the baseline behavior model. A score normalization scheme allows identification of a global threshold to flag current anomalous activity in the user population.

Abstract: An exemplary system, method, and computer-accessible medium can include, for example, receiving an original dataset(s), receiving a synthetic dataset(s), training a model(s) using the original dataset(s) and the synthetic dataset(s), and evaluating the synthetic dataset(s) based on the training of the model(s). The model(s) can include a first model and a second model, and the first model can be trained using the original dataset(s) and the second model can be trained using the synthetic dataset(s). The synthetic dataset(s) can be evaluated by comparing first results from the training of the first model to second results from the training of the second model.