Abstract: Aspects of the present disclosure relate to watermarks and watermarking techniques for data streaming pipelines. Time stamp and offset timeline data is shared by computing instances along the pipeline to enable improved watermarking of the data stream through the pipeline. The improved watermarks enable better determination of completeness for the data stream and improve materialization of the results. The watermarking techniques can include periodically publishing watermark data by processing units of a vertex, fetching a merged watermark for a vertex by a vertex, and/or watching a data storage for the watermark data for events. Consensus algorithms can be used to maintain consensus among vertices for the watermark data.

Abstract: A method may include obtaining a knowledge graph including entities, and determining, for the knowledge graph, a first state including a first selectable entity subset of the entities that are selectable by a user. The first selectable entity subset may include an entity. The method may further include receiving, from the user and via a graphical user interface (GUI), a selection of the entity from the first selectable entity subset, and responsive to the selection, adding the entity to a report schema. The method further includes restricting, by the GUI, a space of selectable entities addable to the knowledge graph to the first selectable entity subset. The report schema may be used to populate a report. The method may further include, responsive to the selection, transitioning the knowledge graph to a second state including a second selectable entity subset of the entities that are selectable by the user.

Abstract: The present disclosure relates to processing support data to increase a self-support knowledge base. In some embodiments, assisted support data is received comprising a record of an interaction between a user and a support professional. In certain embodiments, a support data set is extracted from the assisted support data. In some embodiments, feedback related to the support data set is received. The feedback may include an indication that the support data set is ready to be included in the self-support knowledge base. In some embodiments, upon determining, based on the feedback, that the support data set is ready to be used for self-support, the support data set is added to the self-support knowledge base. The self-support knowledge base may be accessible by a plurality of users.

Abstract: Aspects of the present disclosure provide techniques for improved automated parsing and display of electronic documents. Embodiments include identifying a set of topics in a first electronic document based on one or more rules related to one or more keywords in the first electronic document. Embodiments include providing one or more inputs to a machine learning model based on the set of topics and a second electronic document related to the first electronic document. Embodiments include receiving, from the machine learning model in response to the one or more inputs, one or more outputs related to formatting the second electronic document for display. Embodiments include generating a formatted version of the first electronic document based on the set of topics and generating a formatted version of the second electronic document based on the one or more outputs.

Abstract: Systems and methods for training machine learning models based on domain constraints are disclosed. An example method includes receiving a plurality of images, each image associated with a cluster of a plurality of clusters, the plurality of clusters representing an output of a second machine learning model, assigning a label to each cluster of the plurality of clusters based at least in part on a plurality of constraints, identifying, based at least in part on the plurality of constraints, a first label mismatch for a first image, the first label mismatch indicating that the first image belongs to a first cluster but should be assigned to a second cluster different from the first cluster, reassigning the first image to the second cluster, and training the first machine learning model, based on the labeled clusters of the plurality of clusters, to predict labels associated with subsequently received image data.

Abstract: A method for making binary predictions for a subject involves obtaining historical data for multiple subjects, the historical data including, for each subject, a feature set and a binary outcome, generating training data from the historical data, and training a predictive model using the training data to predict the outcomes based on the feature sets. The method further includes obtaining historical data including a feature set for a subject under consideration, and predicting a binary outcome for the subject under consideration, based on the feature set associated with the subject under consideration.

Abstract: A machine learning system executed by a processor may generate predictions for a variety of natural language processing (NLP) tasks. The machine learning system may include a single deployment implementing a parameter efficient transfer learning architecture. The machine learning system may use adapter layers to dynamically modify a base model to generate a plurality of fine-tuned models. Each fine-tuned model may generate predictions for a specific NLP task. By transferring knowledge from the base model to each fine-tuned model, the ML system achieves a significant reduction in the number of tunable parameters required to generate a fine-tuned NLP model and decreases the fine-tuned model artifact size. Additionally, the ML system reduces training times for fine-tuned NLP models, promotes transfer learning across NLP tasks with lower labeled data volumes, and enables easier and more computationally efficient deployments for multi-task NLP.

Abstract: A method for computer estimations based on statistical tree structures involves obtaining a statistical tree structure for reference elements. The statistical tree structure includes leaf nodes segmenting a statistic for a data label according to data features in the reference elements, and intermediate nodes connecting a first node to the leaf nodes. Each of the first node and the intermediate nodes provide a branching based on one of the data features. The method further includes obtaining target data, including values for the data features, and a value for the data label. The method also includes selecting the first node, associated with a first data feature, traversing the statistical tree structure to a leaf node by matching the values of the data features to the branching of the intermediate nodes, and assessing the value for the data label in the target data based on the statistic associated with the leaf node.

Abstract: Certain aspects of the present disclosure provide techniques for operation of a feature management platform. A feature management platform is an end-to-end platform developed to manage the full lifecycle of data features. For example, to create a feature the feature management platform can receive a processing artifact (e.g., a configuration file and code fragment) from a computing device. The processing artifact defines the feature, including the data source to retrieve event data from, when to retrieve the event data, the type of transform to apply, etc. Based on the processing artifact, the feature management system generates a processing job, which when initiated generates a vector that encapsulates the feature data. The vector is transmitted to the computing device that locally hosts a model, which generates a prediction. The prediction is transmitted to the feature management platform and can be transmitted to other computing devices, upon request.

Abstract: Transactions include text fields, such as description fields. Transactions are extracted from financial institutions using web-scraping extraction. In the process of extracting transactions, errors can be introduced into text fields, such as the inclusion of a space within a word or the removal of a space between words. A statistical approach is applied to the text fields. When two alternative text fields are presented, the alternative that statistically includes more common tokens, such as unigrams and bigrams, is chosen as the correct alternative. The incorrect alternative is replaced by the correct alternative in the text field.

Abstract: Systems and methods for predicting account action failures are disclosed. An example method is performed by one or more processors of a system and includes detecting a character sequence being entered for identifying an account to be used in an account action, extracting a set of subsequences from the character sequence, determining, for the character sequence and each subsequence extracted from the character sequence, a feature value for each of a predefined set of features for characterizing sequences, the determining based on a number of characters included in at least one portion of the character sequence, and generating, using the feature values as input to a model trained in conjunction with a machine learning (ML) algorithm, a predictive value suggestive of a likelihood that the detected character sequence, if submitted, will result in the account action failing.

Abstract: A method may include receiving, for a package, shipment details including attributes, obtaining, for a subset of the attributes, logistic preferences, applying the logistic preferences to the shipment details to obtain modified shipment details, training a classifier using shipment transactions each including values for the attributes and labeled with a vendor logistic service, generating, by applying the classifier to the modified shipment details, scores for vendor logistic services, and recommending a vendor logistic service from the vendor logistic services using the scores.

Abstract: Aspects of the present disclosure provide techniques for document classification through signal processing. Embodiments include receiving a document for classification. Embodiments include generating an image of the document. Embodiments include producing a signal representation of the document based on numbers of non-white pixels in each horizontal scan line or vertical scan line of the image of the document. Embodiments include comparing the signal representation of the document to signal representations of previously-classified documents. Embodiments include determining, based on the comparing, a classification for the document. Embodiments include performing additional processing with respect to the document based on the classification for the document.

Abstract: Certain aspects of the present disclosure provide techniques and systems for automatically detecting, tracking, and processing certain information content, based on voice input from a user. A voice enabled content tracking system receives natural language content corresponding to audio input from a user. A determination is made as to whether the natural language content includes a first type of information, based on evaluating the natural language content with a first machine learning model. In response to determining the natural language content comprises the first type of information, a temporal association of the first type of information is determined, based on evaluating the natural language content with a second machine learning model, and a message including an indication of the temporal association of the first type of information is transmitted to the user.

Abstract: Systems and methods for selectively updating permissions associated with a cloud resource deployment are disclosed. An example method includes receiving a first request to deploy first target cloud resources based on a first specified state defined in a configuration repository, selectively updating deployment permissions associated with the first specified state and deploying the first target cloud resources based at least in part on the first specified state and the updated deployment permissions.

Abstract: Systems and methods are disclosed for model based document image enhancement. Instead of requiring paired dirty and clean images for training a model to clean document images (which may cause privacy concerns), two models are trained on the unpaired images such that only the dirty images are accessed or only the clean images are accessed at one time. One model is a first implicit model to translate the dirty images from a source space to a latent space, and the other model is a second implicit model to translate the images from the latent space to clean images in a target space. The second implicit model is trained based on translating electronic document images in the target space to the latent space. In some implementations, the implicit models are diffusion models, such as denoising diffusion implicit models based on solving ordinary differential equations.

Abstract: A method may include selecting a cohort of entities for migration from a source storage repository to a target storage repository, obtaining a mapping between a source storage schema of the source storage repository to a target storage schema of the target storage repository, and migrating data for the entities in the cohort. Migrating the data of an entity may include copying, without locking the data in the source storage repository and in the target storage repository, the data from the source storage repository to the target storage repository, verifying, while the data is locked, that the data in the source storage repository is the same as the data in the target storage repository, changing, while the data in the source storage repository and the target storage repository is locked, an entity pointer for the entity to the target storage repository based on the verifying, and unlocking the data.

Abstract: A method includes receiving a data stream comprising content generated by an application executing on a user device. The data stream is received from a guidance service that is separate from the application. The data stream is processed using a set of machine learning models to identify a first set of artifacts within the content. A first state of the application is identified based on the first set of artifacts. First transition data is identified in a logic flow of the application. The first transition data corresponds to transitioning from the first state to a second state of the application. Based on the first transition data, first guidance data is generated that describes user input for transitioning the application from the first state to the second state. The first guidance data is sent to the user device, where it is separately presented from the application by the guidance service.

Abstract: A method implements hybrid artificial intelligence generated actionable recommendations. The method includes processing an event to identify an action of an event action set. The event includes an event value. The method further includes processing the event action set to generate an objective value, corresponding to the action, and a probability, corresponding to the action, and to form a model action set from the event action set. The method further includes filtering the model action set using action rule data and rule user data to generate a filtered action set. The method further includes processing, using the objective value and the probability, the filtered action set with an optimization controller to generate suggested action sets from which a selected action set is selected. The selected action set corresponds to a combined action value that satisfies the event value.

Abstract: This disclosure relates to systems and methods for providing user content on a social network. In some aspects, the social network receives, over a communications network from a first computing device associated with a first user of the social network, a transmission including a post to be published on the social network. The social network detects, in the post, goods or services sought or inquired about by the first user, and determines a proximity of the first user. The social network identifies one or more other users of the social network located within a geographical area or the proximity associated with the first user, and presents the post only to the one or more identified users of the social network.