Abstract: A method may include clustering form elements into line objects and columns of a table of a structured representation by applying a trained multi-dimensional clustering model to spatial coordinates of the form elements, and assigning a table header line type to a table header line object of the line objects based on a spatial coordinate of the table header line object relative to a spatial coordinate of a topmost table data line object of the line objects, and a determination that a number of columns of the table header line object is within a threshold of a number of columns of the topmost table data line object. The topmost table data line object may be assigned a table data line type. The method may further include presenting the structured representation to a user.

Abstract: Systems and methods use hierarchical models to process conversations. A set of word vectors is processed using a sentence model, resulting in a sentence vector for the conversation message. The sentence vector is modified to include a time value. A set of sentence vectors, representing a time window, is processed using a window model, to generate a window vector for that time window. The window vector is updated to include a count value. A set of window vectors, corresponding to a set of time windows within the conversation, is processed using a conversation model to generate a conversation vector. A sentiment value indicating the sentiment of the conversation is presented, using a processor that generates the sentiment value from the conversation vector.

Abstract: Certain aspects of the present disclosure provide techniques for ranking users based on similarity. One example method generally includes receiving a request to view a list of users similar to a user based on personal information and receiving user data generated based on attribute data and financial data associated with the user comprising value data of the user. The method further includes obtaining clusters comprising value data of a plurality of users based on mapped data of the plurality of users and determining, by comparing the value data of the user data with the value data of the clusters, a cluster of vectors similar to the user data. The method further includes generating, based on a list of user identifiers using a distance function, a ranked list of similar users and transmitting the ranked list of similar users to a user device for display.

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: Certain aspects of the present disclosure provide techniques for mapping natural language to stored information. The method generally includes receiving a long-tail query comprising a natural language utterance from a user of an application associated with a set of topics and providing the natural language utterance to a natural language model configured to identify nodes of a knowledge graph. The method further includes, based on output of the natural language model, identifying a node of a knowledge graph associated with the natural language utterance, wherein the output of the natural language model includes a node identifier for the node of the knowledge graph and providing the node identifier to the knowledge engine. The method further includes receiving a response associated with the node of the knowledge graph from the knowledge engine and transmitting the response to the user in response to the long-tail query.

Abstract: Certain aspects of the disclosure pertain to predicting a candidate entity match for a transaction with a machine learning model. A description of a transaction associated with an organization is received as input. In response, at least one machine learning model can be invoked to infer a transaction embedding based on the description, a first score that captures similarity between the transaction embedding entity embeddings associated with a global list of entities and organizations, a second score that captures a probability of interaction between the first organization and the entities based on organization and entity embeddings that capture profile data associated with the organization and the entities, and at least one candidate entity based on the first score and the second score. Finally, the inferred candidate entity can be output for use by an automated data entry or other process or system.

Abstract: A method may include extracting, from a transaction, a routing transit number (RTN) and an account number, embedding the RTN to obtain an RTN vector, embedding the account number to obtain an account number matrix, combining, using a trained machine learning model, the RTN vector and the account number matrix to obtain a combined matrix, and classifying the account number as invalid. The classifying may include applying the trained machine learning model to the combined matrix.

Abstract: A method and computer system are provided for generating a text summary. An input text is processed by a model that comprises a set of attention heads. The model is trained to generate abstractive summaries of text documents. A subset of the attention heads are identified. For each attention head in the subset, a portion is identified from the input text that is used to generate the abstractive summary. For each sentence of the input text, a fractional size is calculated for an intersection of the portion and the respective sentence relative to the respective sentence. A subset of the sentences is then determined, where the respective fractional size of each sentence in the subset meets a first threshold. An extractive summary of the input text is then generated from the subset of sentences.

Abstract: Certain aspects of the disclosure provide systems and methods for receiving a request to process a workload on a remote processing system; determining one or more workload requirements associated with processing the workload; and processing the one or more workload requirements with a contextual bandit machine learning model to generate a processing configuration for the remote processing system. The remote processing system provisions resources based on the processing configuration and processes the workload.

Abstract: A method including transcribing, automatically, an ongoing stream of voice data into text phrases. The method also includes receiving an indication of a selected text phrase in the text phrases. The method also includes converting the selected text phrase to a selected phrase vector. The method also includes generating a subsequent text phrase, after the selected text phrase, from the ongoing stream of voice data, and adding the subsequent text phrase to the text phrases. The method also includes converting the subsequent text phrase to a subsequent phrase vector. The method also includes generating a similarity confidence score from the selected phrase vector and the subsequent phrase vector, using a machine learning model. The method also includes highlighting, responsive to the similarity confidence score exceeding a threshold value, the subsequent text phrase in the text phrases.

Abstract: Systems and methods for updating a data catalog in real-time with changes to metadata are disclosed. An example method may be performed by one or more processors of a system and include detecting changes to metadata in a metadata store, generating an event queue indicating a plurality of change events associated with system objects based on the changes, each of the plurality of change events indicating a number of event attributes and a system object associated with the change event, identifying recent events among the plurality of change events, identifying relevant events among the recent events, extracting unique identifiers from the relevant events, identifying priority objects among the system objects, obtaining, from the metadata store, current metadata for each of the priority objects, and updating information associated with the priority objects in a data catalog, the updating occurring in real-time with the changes to the metadata in the metadata store.

Abstract: A method including preprocessing natural language text by cleaning and vectorizing the natural language text. A first machine learning model (MLM) extracts negative reviews. A first input to the first MLM is the natural language text and a first output of the first MLM is first probabilities that the negative reviews have negative sentiments. The method also includes categorizing the negative reviews by executing a second MLM. A second input to the second MLM is the negative reviews. A second output of the second MLM is second probabilities that the negative reviews are assigned to categories. The method also includes identifying, using a name recognition controller and based on categorizing, a name of a software application in the negative reviews and sorting the negative reviews into a subset of negative reviews relating to the name. The software application is adjusted based on the subset of negative reviews.

Abstract: Systems and methods for training machine learning models are disclosed. An example method includes receiving a semi-labeled set of training samples including a first set of training samples, where each training sample in the first set is assigned a known label, and a second set of training samples, where each training sample in the second set has an unknown label, determining a first loss component, the first loss component providing a loss associated with the first set, determining a second loss component, the second loss component having a value which increases based on a difference between a distribution of individually predicted values of at least the second set and an expected overall distribution of at least the second set, and training the machine learning model, based on the first loss component and the second loss component, to predict labels for unlabeled input data.

Abstract: Certain aspects of the present disclosure provide techniques for performing finite rank deep kernel learning. In one example, a method for performing finite rank deep kernel learning includes receiving a training dataset; forming a set of embeddings by subjecting the training dataset to a deep neural network; forming, from the set of embeddings, a plurality of dot kernels; linearly combining the plurality of dot kernels to form a composite kernel for a Gaussian process; receiving live data from an application; and predicting a plurality of values and a plurality of uncertainties associated with the plurality of values simultaneously using the composite kernel.

Abstract: A method including transforming metrics, related to a computer network environment, into a digital image including pixels that represent the metrics. The computer network environment initially is load balanced by a first load balancing scheme selected from among load balancing schemes. The method also includes generating a classification of the digital image. The method also includes selecting, based on the classification of the digital image, a selected load balancing scheme from among the load balancing schemes. The method also includes changing the first load balancing scheme to the selected load balancing scheme such that the selected load balancing scheme is applied to the computer network environment.

Abstract: A computing system generates a plurality of training data sets for generating the NLP model. The computing system trains a teacher network to extract and classify tokens from a document. The training includes a pre-training stage where the teacher network is trained to classify generic data in the plurality of training data sets and a fine-tuning stage where the teacher network is trained to classify targeted data in the plurality of training data sets. The computing system trains a student network to extract and classify tokens from a document by distilling knowledge learned by the teacher network during the fine-tuning stage from the teacher network to the student network. The computing system outputs the NLP model based on the training. The computing system causes the NLP model to be deployed in a remote computing environment.

Abstract: Certain aspects of the present disclosure provide techniques for configuring a software application through a remote configuration service. An example method generally includes receiving, from a remote configuration service, a declarative construct. Generally, the declarative construct includes a definition of a workflow in an application to be executed within a player application deployed on a client device. Information associated with the definition of the workflow is extracted by parsing the declarative construct according to a schema defining a format of the declarative construct. The workflow is executed in the player application based on the extracted information defining functionality of the workflow.

Abstract: A method implements efficient counterfactual search. The method includes receiving a request corresponding to an input vector, processing the input vector with a model to generate an output vector that does not correspond to a selected class, and processing the input vector using a component, of a plurality of components, to generate a counterfactual vector to the selected class. The plurality of components includes a number of dimensions that is less than a number of features of the input vector. The method further includes processing the counterfactual vector to generate a recommendation and presenting the recommendation.