Dell Products Inc.

Dell & Products L.p.

Saurabh Jha

Atul Kumar

Sailendu Kumar Patra

Kuruba Ajay Kumar

Priya Shanmugasundaram

Anamika Chatterjee

Srishti Gupta

Ajay Kumar Kuruba

Kailash Talreja

Rohitashwa Chakraborty

Shweta Goyal

Adya Jha

A method for training a model to generate a structured query language (SQL) output using a natural language input includes receiving a training data set that includes natural language queries, corresponding SQL statements, and a corresponding database. The method also includes extracting query-specific schema from the corresponding database. Further, the method includes embedding the natural language queries and the query-specific schema to generate initialized embeddings, where each initialized embedding corresponds to a natural language query of the natural language queries, and where the corresponding SQL statement is one of the corresponding SQL statements; generating subgraphs based on the initialized embeddings and the natural language queries; training a graph neural network (GNN) using the subgraphs as inputs; generating, using the GNN, refined outputs based on the subgraphs; and training a transformer model using the refined outputs and the natural language queries to obtain a trained model.

Contextual graph dynamics for natural language to SQL conversion

Techniques described herein relate to a method for performing knowledge extraction and noise removal for prediction models. The method includes obtaining, by a prediction system, live tabular data; in response to obtaining live tabular data: performing data preprocessing on the live tabular data to generate processed live tabular data; generating a knowledge vector based on the processed live tabular data using a dimensionality reduction model and a tabular attention model; generating a prediction using a prediction model and the knowledge vector; and providing the prediction to a client; wherein the client performs prediction processing using the prediction.

Method and system for performing noise removal and knowledge extraction to improve prediction model performance

A method for extracting text information from images includes obtaining an extraction request associated with live data comprising an image; generating, using a prediction model, rotational variant features and rotational invariant features associated with the live data; generating, using the prediction model, text embeddings associated with the rotational variant features using overlapping kernel-based embedding on the live data; generating, using the prediction model, attention values for each pixel in the live data using context attention; applying a trained language model to the text embeddings, attention values, and the live data to generate predictions; and performing extraction actions based on the predictions.

Text-based information extraction from images

Document retrieval with domain adaptation and out of distribution identification is disclosed. Custom layers are added to a model without impacting base weights. The custom layers are trained using a domain specific dataset and topic identifiers. When executing a query, the custom layers use weights associated with a topic identifier. If a potential result is sufficiently similar, the result is returned in response to the query. Otherwise, the result is out of distribution and no result is returned.

System and method for document retrieval with domain adaption and out of distribution identification

Methods, apparatus, and processor-readable storage media for localized anomaly detection in digital documents are provided herein. An example computer-implemented method includes identifying at least one image in a digital document; processing the identified at least one image using at least an image compression algorithm; applying a first machine learning model to the at least one processed image, wherein the first machine learning model is trained to detect whether the at least one image comprises one or more modifications; in response to detecting that the at least one image comprises at least one modification, applying a second machine learning model to identify a location in the at least one image corresponding to the at least one modification; and generating an indication that identifies the location of the at least one modification in the at least one image.

Localized anomaly detection in digital documents using machine learning techniques

A method for automatically responding to a user input includes receiving the user input. The method also includes identifying a current domain associated with the user input. Further, the method includes determining, using a previously trained learning model, a first belief state, where the first belief state is based on the current domain. In addition, the method includes determining, using a reinforcement learning model, a second belief state, where the second belief state is based on the current domain and a reward information. Moreover, the method includes determining an action based on the user input and one selected from the group consisting of the first belief state and the second belief state. Also, the method includes generating a response based on the action and presenting the response to a user.

Growing community of inventors

Austin, TX, United States of America

Saurabh Jha