Abstract: A method includes receiving, by a controller, a set of data. The set of data includes data of at least one explanatory variable and data of at least one response variable. The method also includes receiving, by the controller, an input that identifies a relevant explanatory variable. The method also includes generating, by the controller, an output data that corresponds to a symbolic expression that represents a relationship between the at least one explanatory variable and the at least one response variable. The symbolic expression is responsive to the identified relevant explanatory variable, and the symbolic expression includes a mathematically stable expression.

Abstract: Techniques regarding quantum algorithm concatenation are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise a concatenation component, operatively coupled to the processor, that can concatenate a first quantum algorithm and a second quantum algorithm by using an output of the first quantum algorithm as an initial parameter in the second quantum algorithm.

Abstract: Fabric validation using spectral measurement is provided. In various embodiments, a near-infrared absorption spectrum of a fabric sample is received from a near-infrared spectrometer. A plurality of features is extracted from the spectrum. The plurality of features is provided to a trained classifier. The trained classifier provides a similarity score indicative of the similarity of the fabric sample to a reference fabric sample.

Abstract: The disclosed embodiments provide a system for processing data. During operation, the system obtains a set of feature additions and an evaluation metric for assessing the performance of a statistical model. Next, the system automatically builds treatment versions of the statistical model using a set of baseline features for the statistical model and feature combinations generated using the feature additions. The system then uses a hypothesis test and a fixed set of feature values to compare a baseline value of the evaluation metric for a baseline version of the statistical model that is built using the set of baseline features with additional values of the evaluation metric for the treatment versions. Finally, the system outputs a result of the hypothesis test for use in assessing an impact of the feature combinations on a performance of the statistical model.

Abstract: A facility for identifying combinations of feature and machine learning algorithm parameters, where each combination can be combined with one or more machine learning algorithms to train a model, is disclosed. The facility evaluates each genome based on the ability of a model trained using that genome and a machine learning algorithm to produce accurate results when applied to a validation data set by, for example, generating a fitness or validation score for the trained model and the corresponding genome used to train the model. Genomes that produce fitness scores that exceed a fitness threshold are selected for mutation, mutated, and the process is repeated. These trained models can then be applied to new data to generate predictions for the underlying subject matter.

Abstract: Methods and apparatus are described by which artificial intelligence (AI) is used to enable the rapid development of reliable test suites for web and mobile applications. An AI agent guided by reinforcement learning explores an application-under-test (AUT), interacting with the AUT to traverse the flows through the AUT by seeking novel application states. A subset of these flows is then identified as being representative of the functionality of the AUT. The interactions between the AI agent and the AUT that define these identified flows form the basis for the test suite.

Abstract: An example apparatus includes a feature extractor to generate a first image descriptor based on a first image of a first retail product tag corresponding to a first category, the first image descriptor representative of one or more visual features of the first retail product tag; a feature descriptor generator to generate a feature descriptor corresponding to the first retail product tag by concatenating the first image descriptor and a first category signature corresponding to the first retailer category; and a classifier to generate a first probability value corresponding to a first type of promotional product tag and a second probability value corresponding to a second type of promotional product tag based on the feature descriptor; and determine whether the first retail product tag corresponds to the first type of promotional product tag or the second type of promotional product tag based on the first and second probability values.

Abstract: A computer-implemented method is provided for statistical modeling of entities of a particular type. The method can include obtaining entity data including a plurality of entity data sets, each entity data set associated with a respective entity and including values for one or more static parameters indicative of a type of the entity. Each entity data set can include (i) values for input parameter(s) indicative of a security profile of the entity and (ii) a value of a security class parameter indicative of a security class of the entity based on the values of the input parameters. The method can include training a statistical classifier to infer a value of the security class parameter indicative of the security class of a particular entity of the particular type based on values of one or more of the input parameters indicative of a security profile of the particular entity.

Abstract: A method and an invoice analytics system (IAS) for analyzing an entity's business service transactional invoice data (BSTID) and performing invoice analytics (IA) in real time are provided. The IAS extracts contract line item (CLI) data from contracts and stores the CLI data in a CLI database. The IAS transforms the CLI data into a quick view format and generates a contract quick view interface that displays the CLI data on an entity device. The IAS extracts and segments invoice line item (ILI) data from aggregated BSTID, stores the ILI data in an ILI database, and reconciles the invoices in accordance with the contracts by identifying and rectifying errors and off-contract business service items. The IAS performs IA on the ILI data for computing cost savings and generates an interactive, dynamic, and searchable IA report including graphical data representations of purchase patterns, outliers, variations, and predictive analytics for the entity.

Abstract: Systems and methods for fraud containment are provided in accordance with an embodiment of the invention. A fraud event may be detected. One or more devices that may be used in perpetrating the fraud event may be detected. Additionally one or more potential fraud victims, who may be grouped into victim circles may be detected. The threat level to the victims and/or victim circles may be assessed. In some instances, behavioral profiles may be utilized to make fraud assessments. Based on the threat level, recommendations for fraud containment responses may be provided.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying a paddle region. One of these method includes generating a map of a paddle region for a paddler of an aquatic device based on weather, nautical conditions, experience level of the paddler, and the duration of the trip; and presenting the map to the user on a computer device.

Abstract: In some implementations, a method performed by data processing apparatuses includes identifying a first delivery route that includes delivery stops at multiple locations within a first geographic region, and identifying a second delivery route within a second geographic region. The first delivery route can be assigned to a first carrier that is designated to deliver orders within the first geographic region, and the second delivery route can be assigned to a second carrier that is designated to deliver orders within the second geographic region. An additional order can then be identified that is requested to be delivered to a location within the first geographic region. The second delivery route can be modified to include the additional delivery stop for the additional order, including assigning the additional order to the second carrier even though the second carrier is designated to deliver orders within the second geographic region.

Abstract: A computer implemented method for incorporating multiple objectives in a ranked list of search results includes receiving a search query from a client device, accessing a set of stored listings for goods or services and probabilities of serving the listings, defining a serving vector as a probability distribution over the set of listings, providing a serving vector as input to a multi-objective function, decomposing the multi-objective function into one or more objective functions, generating a ranked list of the listings based at least in part on the serving vector that maximizes the decomposed multi-objective function, and providing the listings to the client device according to the order of the ranked list. Each objective function addresses a different goal in an overall diversity optimization.

Abstract: Implementing an application within a heterogeneous device can include receiving an application specifying a plurality of hardware accelerators and having a plurality of sections corresponding to different subsystems of the heterogeneous device, wherein the plurality of sections are specified using different programming models. Compiling each section based on the programming model of the section and the subsystem of the heterogeneous device corresponding to the section into an accelerator representation. Linking the accelerator representations based on a platform of the heterogeneous device, generating a hardware implementation of the application for the heterogeneous device based on the linked accelerator implementations, and automatically generating program code configured to control one or more of the plurality of hardware accelerators of the hardware implementation.

Abstract: A method evolves generic computational building blocks. The method initializes a parent population with randomly generated programs. The method also obtains a list of randomly generated test inputs. The method generates a target dataset that includes input-output pairs of randomly generated binary strings. The method also applies a fitness function to assign a fitness score to each program, based on the input-output pairs of the target dataset. The method grows a seed list by applying genetic operators to the programs, and selecting offspring that satisfy a novelty condition. The novelty condition is representative of an ability of a program to produce distinct and unique output for the list of randomly generated test inputs. The method iterates until a terminating condition has been satisfied. The terminating condition is representative of an ability of programs in the seed list to solve one or more genetic programming instances.

Abstract: A device may receive historical software data associated with a software development platform, and may train a neural network model, with the historical software data, to generate a trained neural network model. The device may receive, from the software development platform, historical test configuration data, and may train an execution model, with the historical test configuration data, to generate a trained execution model. The device may receive software data identifying software to be tested, and may process the software data, with the trained neural network model, to predict a set of test cases to execute for testing the software. The device may process the set of test cases, with the trained execution model, to identify configurations, scripts, and test targets for executing the set of test cases, and may perform one or more actions based on the set of test cases, the configurations, the scripts, and the test targets.

Abstract: Techniques to facilitate generation of controller application code that emulates functionality of an industrial controller are disclosed herein. In at least one implementation, a computing system interfaces with the industrial controller and monitors input and output states of the industrial controller while the industrial controller operates a machine system. The input and output states of the industrial controller used to operate the machine system are analyzed to generate a functional design specification for the industrial controller. The controller application code that emulates the functionality of the industrial controller is generated based on the functional design specification.

Abstract: A framework for generating optimized recommendations is described herein. For example, an optimized customer recommendation engine is described herein. Customer data is collected and pre-processed into a data model. Recommendations are calculated and provided by an aggregated method. The aggregated output is generated based on the outputs of a real-time prediction model and an offline modeling process. The real-time prediction model may be an online modeling training technique based on support vector machines (SVM) to classify customers and provide quick recommendations. The offline modeling process may be a learning process based on a back-propagation artificial neural network (BP-ANN) to provide with reliable predictions. Validation may be introduced to evaluate the accuracy of the recommendation model.

Abstract: An example method may include a processing system including at least one processor determining a final weight set comprising weight factors to apply to each of a plurality of performance indicators for a predictive model associated with a target performance indicator using a genetic algorithm. The method may further include the processing system gathering a first plurality of measurements of the plurality of performance indicators for at least a portion of a cellular network, applying the predictive model to the first plurality of measurements of the plurality of performance indicators to generate a prediction for the target performance indicator, and adjusting at least one aspect of the cellular network in response to the prediction.

Abstract: A method of providing reservation status relating to a group of objects includes providing a database for containing (i) availability data describing availability of at least one of the group of objects, and (ii) reservation data describing allocation of at least one of the group of objects. The availability data and the reservation data are stored in sparse form. The method further includes extracting, from the database, availability data and reservation data corresponding to a predetermined period of time, and combining the extracted availability data and reservation data to form windowed data in a dense format. The method also includes receiving the windowed data and providing the windowed data in a bit-vector view, and sampling the windowed data in the bit-vector view to produce sampled data. The sampled data provides the reservation status as a static view of the windowed data in the bit-vector view.

Abstract: A method and associated systems for using machine-learning methods to perform linear regression on a DNA-computing platform. One or more processors generate and initialize beta coefficients of a system of linear equations. These initial values are encoded into nucleobase chains that are then padded to a standard length. The chains are allowed to bind with complementary template chains in a DNA-computing reaction, and the resulting DNA molecules are decoded to reveal the relative the relative likelihood of each chain to bind. The initial values of the beta coefficients are weighted proportionally to these likelihoods, and the process is repeated iteratively until the beta coefficients converge to optimal values.

Abstract: In accordance with some embodiments, a training method for biometric identity and authentication is provided. The method includes obtaining biometric data from a plurality of sources. The method further includes extracting a plurality of feature vectors from the biometric data. The method also includes determining a plurality of identifiability scores correspondingly associated with the plurality of feature vectors, where each of the plurality of identifiability scores provides a quantitative characterization of a relative uniqueness of a corresponding one of the plurality of feature vectors.

Abstract: A method and apparatus of a device that incorporates a user preference into a multi-objective optimization while the multi-objective optimization is running is described. In an exemplary embodiment, the device generates a first plurality of intermediate designs based on optimizing a plurality of variable values corresponding to the multiple dimensions of the multi-objective optimization. In addition, each of the first plurality of intermediate designs includes a corresponding weight. Furthermore, the device outputs the first plurality of intermediate designs. The device additionally receives a preference indicator for a selected one of the intermediate designs, where a user inputs the preference indicator while the multi-objective optimization is running. The device further adjusts the corresponding weight of the selected one of the first plurality of intermediate designs based on the preference indicator.

Abstract: A computer-implemented intent classification method and system are provided. The method comprises: providing a plurality of trained model sets, each based on a different model and comprising a plurality of trained models, with each trained model in each trained model set obtained by training a same model with a different initialization condition over a pre-determined dataset; feeding an input text into each trained model to obtain a plurality of prediction results for indicating a likely intent class of the input text; ensembling prediction results corresponding to a same trained model set for each trained model set to obtain a plurality of first-layer ensembles; ensembling the plurality of first-layer ensembles to obtain at least one second-layer ensemble, each obtained by ensembling two or more first-layer ensembles; and determining the intent class of the input text based on the at least one second-layer ensemble.

Abstract: A method for providing bidding support to a user during an auction using cognitive modeling comprises obtaining data associated with the auction, with the obtained data comprising data associated with at least one bidder. The obtained data is analyzed, and at least one output is generated based on the analysis. The at least one output is sent to at least one device associated with the user. The at least one device is configured to generate at least one alert based on the at least one output, with the at least one alert providing decision support to the user with respect to bidding strategy.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for obtaining an electrocardiographic (ECG) signal of a user; obtaining a feature vector of the ECG signal of the user with neural network based feature extraction. Comparing the feature vector of the ECG signal with a stored feature vector of a registered user. Authenticating the user in response to determining that a similarity of the ECG feature vector of the ECG signal and the stored ECG feature vector of the registered user exceeds a pre-defined threshold value.

Abstract: An algorithm for performing an image or video processing task is generated that may be used to combine a plurality of different independent solutions to the image or video processing task in an optimized manner. A plurality of base algorithms may be applied to a training set of images or video and a first generation of different combining algorithms may be applied to combine the respective solutions from each of the respective base algorithms into respective combined solutions. The respective combined solutions may be evaluated to generate respective fitness scores representing measures of how well the plurality of different combining algorithms each perform the image or video processing task. The algorithms may be iteratively updated to generate an optimized combining algorithm that may be applied to an input image or video.

Abstract: Methods, systems, and computer programs encoded on computer storage medium, for receiving, by an analytics computing engine, a first remote control detail record (RCDR) of a remote control device that is communicatively coupled to a set-top box, the first RCDR including sensor data from one or more sensors of the remote control device, the RCDR being generated in response to a first video stream provided by the set-top box; processing, by the analytics computing engine, the first RCDR to determine a classification of a user associated with the sensor data, the classification of the user including one or more demographic categories; determining, by the analytics computing engine, a second video stream based on the classification of the user; and communicating, by the analytics computing engine, the second video stream to the set-top box.

Abstract: An optimization approach for placement heads tasks assignment of beam-type single-gantry surface mounter is disclosed, building a comprehensive framework that simultaneously achieves all optimization objectives in reasonably short time. Specific steps are as follows: (a) importing user-defined production parameters and forming raw production data, (b) pre-processing the raw production data and obtaining the intermediate production data that facilitates the subsequent processing, (c) verifying the loop condition of the “placement heads tasks assignment”, (d) performing nozzle assignment, (e) confirming all the nozzle assignment exchange possibilities, (f) verifying whether the loop condition of “component allocation” are fulfilled, (g) performing component allocation, and (h) outputting the optimal placement heads tasks assignment results.

Abstract: Features of multiple dimensions are extracted from information included in a URL access request. A risk score of the URL access request is obtained by providing the features to a predetermined URL attack detection model for prediction calculation, where the predetermined URL attack detection model is a machine learning model obtained through training based on the Isolation Forest machine learning algorithm. It is determined, based on the risk score, that the URL access request is a URL attack request.

Abstract: After beginning an evaluation of a first generation of a first plurality of candidate solutions and prior to completion of the evaluation, a system can determine a first plurality of tournament participants from the first plurality of candidate solutions The system may determine whether each of the first plurality of tournament participants have been evaluated. In response to determining each of the first plurality of tournament participants have been evaluated the system can determine a first one or more winners of the tournament and determine whether a first quantity of the first one or more winners is above a predefined threshold. In response to determining the first quantity is above a predefined threshold, the system can determine a second generation of a second plurality of candidate solutions, the second plurality of candidate solutions based, at least in part, on the first one or more winners.

Abstract: A method and apparatus used for general purpose problem solving using entanglement properties of holography. Intelligent point-based entities having spatial and other electromagnetic properties called DROPLETS [Data-Representative-Object-Particle(s)-Liking-EnTanglement] are generated as avatars, or delegate objects, connected to concrete or abstract data sources representing a situation, event or other problem. Each DROPLET's properties are controlled by changes in the input sources, feedback, changes in itself, and/or changes of other DROPLETS. Coherent rays are introduced and interact with said DROPLETS, generating an INTELLIGENCE WAVEFRONT. Interference patterns are recorded and converted to binary machine codes used as instruction keys to store and lock human readable and/or machine readable content components into a plurality of associative memories. Said content includes waveforms, harmonics, codes, data, and other holograms.

Abstract: The present invention relates to a computer implemented system for animating a virtual object or digital entity. It has particular relevance to animation using biologically based models, or behavioural models particularly neurobehavioural models. There is provided a plurality of modules having a computational element and a graphical element. The modules are arranged in a required structure and have at least one variable and being associated with at least one connector. The connectors link variables between modules across the structure, and the modules together provide a neurobehavioural model. There is also provided a method of controlling a digital entity in response to an external stimulus.

Abstract: Well settings in mature oil field may be configured by generating progressively, geological reservoir models by fitting a set of initial geological reservoir models that generate forecasts of fluid injection and production at a well as a function of time, with a subset of historical well data to produce a set of intermediary geological reservoir models, and fitting the set of intermediary geological reservoir models to a next subset of the historical well data. A set of diverse geological reservoir models may be determined from the geological reservoir models. A mid-term schedule of well settings associated with the well may be generated based on the set of diverse geological reservoir models and an economic model that comprises mid-term estimation of oil sale price and of production costs. The mid-term schedule of well settings may be actuated to control the fluid injection and production at the well.

Abstract: The present disclosure relates to a system, method, and computer program for providing users with notifications of a cashback rewards from a shopping portal using screen and email analysis. A shopping portal system analyzes the content and characteristics of user emails, as well as screens viewed by the user through a client application (e.g., webpages and mobile application screens), to identify probable order-confirmation emails and screens. In response to identifying an order-confirmation email or an order-confirmation screen, the system determines whether a cashback reward should be credited to the user for the order corresponding to the order-confirmation email/screen. In response to an order-confirmation email or screen satisfying the criteria for a cashback reward, the system credits a user account with the cashback reward and notifies the user of the reward.

Abstract: A method includes, during an epoch of a genetic algorithm, determining a fitness value for each of a plurality of autoencoders. The fitness value for an autoencoder indicates reconstruction error responsive to data representing a first operational state of one or more devices. The method includes selecting, based on the fitness values, a subset of autoencoders. The method also includes performing a genetic operation with respect to at least one autoencoder to generate a trainable autoencoder. The method includes training the trainable autoencoder to reduce a loss function value to generate a trained autoencoder. The loss function value is based on reconstruction error of the trainable autoencoder responsive to data representative of a second operational state of the device(s). The method includes adding the trained autoencoder to a population to be provided as input to a subsequent epoch of the genetic algorithm.

Abstract: The present invention provides devices, methods, and systems for coupling a rational agent to a quantum process. In particular, the present invention provides rational agents configured to influence a quantum process, or to derive information from a quantum process, and methods and uses thereof.

Abstract: Roughly described, anomalous behavior of a machine-learned computer-implemented individual can be detected while operating in a production environment. A population of individuals is represented in a computer storage medium, each individual identifying actions to assert in dependence upon input data. As part of machine learning, the individuals are tested against samples of training data and the actions they assert are recorded in a behavior repository. The behavior of an individual is characterized from the observations recorded during training. In a production environment, the individuals are operated by applying production input data, and the production behavior of the individual is observed and compared to the behavior of the individual represented in the behavior repository. A determination is made from the comparison of whether the individual's production behavior during operation is anomalous.

Abstract: Various embodiments are described that relate to an adaptive learning system. The adaptive learning system can be trained by correlation between a first set of raw technical performance data and a set of actual operational effectiveness assessment data. Once trained, the adaptive learning system can be deployed. Once deployed, the adaptive learning system can produce a set of predicted operational effectiveness assessment data from a second set of raw technical performance data that is different from the first set of raw technical performance data.

Abstract: A method for resolving, by a mobile device, resource collision in a V2X communication environment may comprise the steps of: determining whether resource collision has occurred in a specific resource selected by a plurality of adjacent mobile devices; and, if it is determined that resource collision has occurred, transmitting information on the result of the determination to the plurality of adjacent mobile devices.

Abstract: A computer-implemented method is provided for statistical modeling of entities of a particular type. The method can include obtaining entity data including a plurality of entity data sets, each entity data set associated with a respective entity and including values for one or more static parameters indicative of a type of the entity. Each entity data set can include (i) values for input parameter(s) indicative of a security profile of the entity and (ii) a value of a security class parameter indicative of a security class of the entity based on the values of the input parameters. The method can include training a statistical classifier to infer a value of the security class parameter indicative of the security class of a particular entity of the particular type based on values of one or more of the input parameters indicative of a security profile of the particular entity.

Abstract: There are provided measures for machine learning based malware detection systems. Such measures exemplarily include analyzing a set of training data, said set of training data comprising a plurality of training data elements, wherein each of said plurality of training data elements is associated with a respective one of at least two maliciousness related properties, learning a malicious object detection model on the basis of first feature combinations of said plurality of training data elements, said first feature combinations characterizing each of said at least two maliciousness related properties, learning an anomalous data detection model on the basis of second feature combinations of said plurality of training data elements, said second feature combinations characterizing said set of training data, said anomalous data detection model being associated with said malicious object detection model, and providing said malicious object detection model and said anomalous data detection model.

Abstract: Methods and systems for model discovery include forming a mathematical program based on a set of observational data to generate an objective function and one or more constraints. The mathematical program represents a model space as an expression tree comprising operators and operands. The mathematical program is solved by optimizing the objective function subject to the one or more constraints to determine a model in the model space that best fits the set of observational data.

Abstract: Techniques are provided for determining labels associated with first and second candidate code and whether those labels are compatible. When the first candidate code and the second candidate code are compatible, third candidate code based is determined on the those two. When the third candidate code meets exit criteria the third candidate code is provided as evolved code. Some embodiments also include causing execution of the evolved code.

Abstract: Systems and methods for storing a plurality of conversations are provided. Each such conversation has a corresponding conversation identifier and one or more messages sharing a common set of characteristics that meet predefined criteria. Information is sent to a remote computer identifying the conversations and providing sufficient information to rank them in accordance with a date/time value of a most recently received message of each conversation thereby causing a first conversation to be ordered before a second conversation. Subsequently, a first message that is deemed to belong to the second conversation is obtained and added to the second conversation. Information is resent to a remote computer identifying the plurality of conversations and providing sufficient information to rank the plurality of conversations in accordance with a date/time value of a most recently received message of each respective conversation thereby causing the first conversation to be ordered after the second conversation.

Abstract: A method, system, and apparatus for managing a part design. The method comprises processing, by a computer system, a finite element model for a part to refine a part topology and generate a number of configurations, in which each configuration has a part weight and a number of different parameters that change the part weight and meet a specification for the part. The method analyzes a part geometry model and a selected manufacturing process to estimate a resource use for manufacturing the part. In response to determining that the part weight or the resource use for a configuration is unacceptable, the method iteratively edits the configuration, refines the part topology to generate an updated configuration, and analyzes the selected manufacturing process to estimate the resource use, until an optimized configuration is identified. The method outputs a file for the optimized configuration, enabling manufacturing of the part using the part design.

Abstract: A device may obtain, for a set of transactions, a set of transaction values associated with a particular industry. The device may determine one or more sample statistical distributions for a probabilistic transaction model by using one or more machine learning techniques. The one or more sample statistical distributions may be similar to one or more actual statistical distributions that are associated with the set of transaction values. The device may generate simulated transaction information using the probabilistic transaction model. The device may perform one or more actions after generating the simulated transaction information.

Abstract: Feature selection methods and processes that facilitate reduction of model components available for iterative modeling. It has been discovered that methods of eliminating model components that do not meaningfully contribute to a solution can be preliminarily discovered and discarded, thereby dramatically decreasing computational requirements in iterative programming techniques. This development unlocks the ability of iterative modeling to be used to solve complex problems that, in the past, would have required computation time on orders of magnitude too great to be useful.

Abstract: A method of generating a neural network includes iteratively performing operations including generating, for each neural network of a population, a matrix representation. The matrix representation of a particular neural network includes rows of values, where each row corresponds to a set of layers of the particular neural network and each value specifies a hyperparameter of the set of layers. The operations also include providing the matrix representations as input to a relative fitness estimator that is trained to generate estimated fitness data for neural networks of the population. The estimated fitness data are based on expected fitness of neural networks predicted by the relative fitness estimator. The operations further include generating, based on the estimated fitness data, a subsequent population of neural networks. The method also includes, when a termination condition is satisfied, outputting data identifying a neural network as a candidate neural network.

Abstract: Techniques are described herein for automatically generating data extraction templates for structured documents (e.g., B2C emails, invoices, bills, invitations, etc.), and for assigning classifications to those data extraction templates to streamline data extraction from subsequent structured documents. In various implementations, a data extraction template generated from a cluster of structured documents that share fixed content may be identified. Features of the cluster of structured documents may be applied as input to extraction machine learning model(s) trained to provide location(s) of transient field(s) in structured documents, to determine location(s) of transient field(s) in the cluster of structured documents. An association between the data extraction template and the determined transient field location(s) may be stored. Based on the association, data point(s) may be extracted from a given structured document of a user that shares fixed content with the cluster of structured documents.