Patents by Inventor Sandeep Tata
Sandeep Tata has filed for patents to protect the following inventions. This listing includes patent applications that are pending as well as patents that have already been granted by the United States Patent and Trademark Office (USPTO).
-
Publication number: 20240126827Abstract: Systems and methods for efficiently identifying and extracting machine-actionable structured data from web documents are provided. The technology employs neural network architectures which process the raw HTML content of a set of seed websites to create transferrable models regarding information of interest. These models can then be applied to the raw HTML of other websites to identify similar information of interest. Data can thus be extracted across multiple websites in a functional, structured form that allows it to be used further by a processing system.Type: ApplicationFiled: December 13, 2023Publication date: April 18, 2024Inventors: Ying Sheng, Yuchen Lin, Sandeep Tata, Nguyen Vo
-
Publication number: 20240054390Abstract: Labels are often over labeled by machine-learning models and under labeled by human labelers. A solution to the over and under labeling problem is to have both a machine-learning model and a human label a document, then send the document to a parser to determine the discrepancies. The discrepancies are then presented to a human to review and decide whether the machine-learning model identified labels are labels. The feedback is then given to the machine-learning model for further improvement in its confidence calculations which via a confidence threshold determine if the identified labels are presented.Type: ApplicationFiled: August 19, 2022Publication date: February 15, 2024Inventors: James Bradley Wendt, Sandeep Tata, Lauro Ivo Beltrao Colaco Costa, Emmanouil Koukoumidis
-
Publication number: 20240046684Abstract: The present disclosure is directed to extracting text from form-like documents. In particular, a computing system can obtain an image of a document that contains a plurality of portions of text. The computing system can extract one or more candidate text portions for each field type included in a target schema. The computing system can generate a respective input feature vector for each candidate for the field type. The computing system can generate a respective candidate embedding for the candidate text portion. The computing system can determine a respective score for each candidate text portion for the field type based at least in part on the respective candidate embedding for the candidate text portion. The computing system can assign one or more of the candidate text portions to the field type based on the respective scores.Type: ApplicationFiled: October 19, 2023Publication date: February 8, 2024Inventors: Sandeep Tata, Bodhisattwa Prasad Majumder, Qi Zhao, James Bradley Wendt, Marc Najork, Navneet Potti
-
Patent number: 11886533Abstract: Systems and methods for efficiently identifying and extracting machine-actionable structured data from web documents are provided. The technology employs neural network architectures which process the raw HTML content of a set of seed websites to create transferable models regarding information of interest. These models can then be applied to the raw HTML of other websites to identify similar information of interest. Data can thus be extracted across multiple websites in a functional, structured form that allows it to be used further by a processing system.Type: GrantFiled: January 29, 2020Date of Patent: January 30, 2024Assignee: Google LLCInventors: Ying Sheng, Yuchen Lin, Sandeep Tata, Nguyen Vo
-
Patent number: 11830269Abstract: The present disclosure is directed to extracting text from form-like documents. In particular, a computing system can obtain an image of a document that contains a plurality of portions of text. The computing system can extract one or more candidate text portions for each field type included in a target schema. The computing system can generate a respective input feature vector for each candidate for the field type. The computing system can generate a respective candidate embedding for the candidate text portion. The computing system can determine a respective score for each candidate text portion for the field type based at least in part on the respective candidate embedding for the candidate text portion. The computing system can assign one or more of the candidate text portions to the field type based on the respective scores.Type: GrantFiled: July 18, 2022Date of Patent: November 28, 2023Assignee: GOOGLE LLCInventors: Sandeep Tata, Bodhisattwa Prasad Majumder, Qi Zhao, James Bradley Wendt, Marc Najork, Navneet Potti
-
Publication number: 20230014465Abstract: Systems and methods for efficiently identifying and extracting machine-actionable structured data from web documents are provided. The technology employs neural network architectures which process the raw HTML content of a set of seed websites to create transferable models regarding information of interest. These models can then be applied to the raw HTML of other websites to identify similar information of interest. Data can thus be extracted across multiple websites in a functional, structured form that allows it to be used further by a processing system.Type: ApplicationFiled: January 29, 2020Publication date: January 19, 2023Inventors: Ying Sheng, Yuchen Lin, Sandeep Tata, Nguyen Vo
-
Patent number: 11526752Abstract: Provided are computing systems and methods directed to active learning and may provide advantages or improvements to active learning applications for skewed data sets. A challenge in training and developing high-quality models for many supervised learning scenarios is obtaining labeled training examples. Provided are systems and methods for active learning on a training dataset that includes both labeled and unlabeled datapoints. In particular, the systems and methods described herein can select (e.g., at each of a number of iterations) a number of the unlabeled datapoints for which labels should be obtained to gain additional labeled datapoints on which to train a machine-learned model (e.g., machine-learned classifier model). Generally, provided are cost-effective methods and systems for selecting data to improve machine-learned models in applications such as the identification of content items in text, images, and/or audio.Type: GrantFiled: January 23, 2020Date of Patent: December 13, 2022Assignee: GOOGLE LLCInventors: Qi Zhao, Abbas Kazerouni, Sandeep Tata, Jing Xie, Marc Najork
-
Publication number: 20220375245Abstract: The present disclosure is directed to extracting text from form-like documents. In particular, a computing system can obtain an image of a document that contains a plurality of portions of text. The computing system can extract one or more candidate text portions for each field type included in a target schema. The computing system can generate a respective input feature vector for each candidate for the field type. The computing system can generate a respective candidate embedding for the candidate text portion. The computing system can determine a respective score for each candidate text portion for the field type based at least in part on the respective candidate embedding for the candidate text portion. The computing system can assign one or more of the candidate text portions to the field type based on the respective scores.Type: ApplicationFiled: July 18, 2022Publication date: November 24, 2022Inventors: Sandeep Tata, Bodhisattwa Prasad Majumder, Qi Zhao, James Bradley Wendt, Marc Najork, Navneet Potti
-
Patent number: 11393233Abstract: The present disclosure is directed to extracting text from form-like documents. In particular, a computing system can obtain an image of a document that contains a plurality of portions of text. The computing system can extract one or more candidate text portions for each field type included in a target schema. The computing system can generate a respective input feature vector for each candidate for the field type. The computing system can generate a respective candidate embedding for the candidate text portion. The computing system can determine a respective score for each candidate text portion for the field type based at least in part on the respective candidate embedding for the candidate text portion. The computing system can assign one or more of the candidate text portions to the field type based on the respective scores.Type: GrantFiled: June 2, 2020Date of Patent: July 19, 2022Assignee: GOOGLE LLCInventors: Sandeep Tata, Bodhisattwa Prasad Majumder, Qi Zhao, James Bradley Wendt, Marc Najork, Navneet Potti
-
Publication number: 20210374395Abstract: The present disclosure is directed to extracting text from form-like documents. In particular, a computing system can obtain an image of a document that contains a plurality of portions of text. The computing system can extract one or more candidate text portions for each field type included in a target schema. The computing system can generate a respective input feature vector for each candidate for the field type. The computing system can generate a respective candidate embedding for the candidate text portion. The computing system can determine a respective score for each candidate text portion for the field type based at least in part on the respective candidate embedding for the candidate text portion. The computing system can assign one or more of the candidate text portions to the field type based on the respective scores.Type: ApplicationFiled: June 2, 2020Publication date: December 2, 2021Inventors: Sandeep Tata, Bodhisattwa Prasad Majumder, Qi Zhao, James Bradley Wendt, Marc Najork, Navneet Potti
-
Publication number: 20210019622Abstract: A user device can send, to a server, a request for a set of documents likely to be opened by a user, determine a client-suggested document to present to the user and a potential motive for the user to open the client-suggested document, receive a suggestion message from the server, the suggestion message including a set of documents likely to be opened by the user and potential motives for the user to open documents in the set of documents, and present, on a display of the user device, visual representations of the client-suggested document, the potential motive for the user to open the client-suggested document, multiple documents included in the set of documents, and the potential motives for the user to open the multiple documents in the set of documents.Type: ApplicationFiled: October 5, 2020Publication date: January 21, 2021Inventors: Alan Green, Cayden Meyer, Julian Gibbons, Alexandre Mah, Divanshu Garg, Reuben Kan, Michael Smith, Sandeep Tata, Alexandrin Popescul
-
Patent number: 10832130Abstract: A user device can send, to a server, a request for a set of documents likely to be opened by a user, determine a client-suggested document to present to the user and a potential motive for the user to open the client-suggested document, receive a suggestion message from the server, the suggestion message including a set of documents likely to be opened by the user and potential motives for the user to open documents in the set of documents, and present, on a display of the user device, visual representations of the client-suggested document, the potential motive for the user to open the client-suggested document, multiple documents included in the set of documents, and the potential motives for the user to open the multiple documents in the set of documents.Type: GrantFiled: April 5, 2017Date of Patent: November 10, 2020Assignee: GOOGLE LLCInventors: Alan Green, Cayden Meyer, Julian Gibbons, Alexandre Mah, Divanshu Garg, Reuben Kan, Michael Smith, Sandeep Tata, Alexandrin Popescul
-
Publication number: 20200250527Abstract: The present disclosure provides computing systems and methods directed to active learning and may provide advantages or improvements to active learning applications for skewed data sets. A challenge in training and developing high-quality models for many supervised learning scenarios is obtaining labeled training examples. This disclosure provides systems and methods for active learning on a training dataset that includes both labeled and unlabeled datapoints. In particular, the systems and methods described herein can select (e.g., at each of a number of iterations) a number of the unlabeled datapoints for which labels should be obtained to gain additional labeled datapoints on which to train a machine-learned model (e.g., machine-learned classifier model). Generally, the disclosure provides cost-effective methods and systems for selecting data to improve machine-learned models in applications such as the identification of content items in text, images, and/or audio.Type: ApplicationFiled: January 23, 2020Publication date: August 6, 2020Inventors: Qi Zhao, Abbas Kazerouni, Sandeep Tata, Jing Xie, Marc Najork
-
Patent number: 10558615Abstract: Augmenting data files in a repository of an append-only file system includes maintaining a companion metadata file for each corresponding data file in a map-reduce system using the append-only file system. Each companion metadata file tracks a logical end-of-file (EOF) for each data file. Global versioning of each companion metadata is maintained. A map-reduce append job is performed for a set of data files using a current global version number for the companion metadata file. The map-reduce job including multiple append tasks. For each successful append job, a logical EOF for each appended file is incremented to a new physical EOF. For each failed append task of the append job, a logical EOF is maintained for each failed append task by not incrementing the logical EOF for each failed append task.Type: GrantFiled: June 30, 2016Date of Patent: February 11, 2020Assignee: International Business Machines CorporationInventor: Sandeep Tata
-
Patent number: 10078681Abstract: There are provided a method for operating multi-node data stores. The method performs storing a data table in a first computing node and stores an index table in a second computing node. The index table provides keys used for accessing data in the first computing node and other multi-node data stores. The method performs operations that update or read the data table accessed from the first computing node and the index table accessed from the second computing node. The operations optimizes between latency in updating or reading the data table and the index table and data consistency maintained between data entries in the data table and data entries pointed by indices in the index table.Type: GrantFiled: December 21, 2015Date of Patent: September 18, 2018Assignee: International Business Machines CorporationInventors: Wei Tan, Sandeep Tata
-
Patent number: 10078682Abstract: There are provided a system and a computer program product for operating multi-node data stores. The system stores a data table in a first computing node and stores an index table in a second computing node. The index table provides keys used for accessing data in the first computing node and other multi-node data stores. The system performs operations that update or read the data table accessed from the first computing node and the index table accessed from the second computing node. The operations optimizes between latency in updating or reading the data table and the index table and data consistency maintained between data entries in the data table and data entries pointed by indices in the index table.Type: GrantFiled: December 21, 2015Date of Patent: September 18, 2018Assignee: International Business Machines CorporationInventors: Wei Tan, Sandeep Tata
-
Publication number: 20180081503Abstract: A user device can send, to a server, a request for a set of documents likely to be opened by a user, determine a client-suggested document to present to the user and a potential motive for the user to open the client-suggested document, receive a suggestion message from the server, the suggestion message including a set of documents likely to be opened by the user and potential motives for the user to open documents in the set of documents, and present, on a display of the user device, visual representations of the client-suggested document, the potential motive for the user to open the client-suggested document, multiple documents included in the set of documents, and the potential motives for the user to open the multiple documents in the set of documents.Type: ApplicationFiled: April 5, 2017Publication date: March 22, 2018Inventors: Alan GREEN, Cayden MEYER, Julian GIBBONS, Alexandre MAH, Divanshu GARG, Reuben KAN, Michael SMITH, Sandeep TATA, Alexandrin POPESCUL
-
Publication number: 20160306799Abstract: Augmenting data files in a repository of an append-only file system includes maintaining a companion metadata file for each corresponding data file in a map-reduce system using the append-only file system. Each companion metadata file tracks a logical end-of-file (EOF) for each data file. Global versioning of each companion metadata is maintained. A map-reduce append job is performed for a set of data files using a current global version number for the companion metadata file. The map-reduce job including multiple append tasks. For each successful append job, a logical EOF for each appended file is incremented to a new physical EOF. For each failed append task of the append job, a logical EOF is maintained for each failed append task by not incrementing the logical EOF for each failed append task.Type: ApplicationFiled: June 30, 2016Publication date: October 20, 2016Inventor: Sandeep Tata
-
Patent number: 9424271Abstract: Augmenting data files in a repository of an append-only file system comprises maintaining metadata corresponding to each data file for tracking a logical end-of-file (EOF) for each data file for appending. A global versioning mechanism for the metadata allows selecting the current version of the metadata to read for performing an append job for a set of data files. Each append job comprises multiple append tasks. For each successful append job, the global versioning mechanism increments a valid metadata version to use for each data file appended. Said valid metadata version indicates the logical EOF corresponding to a new physical EOF for each of the data files appended.Type: GrantFiled: August 30, 2012Date of Patent: August 23, 2016Assignee: International Business Machines CorporationInventor: Sandeep Tata
-
Publication number: 20160188649Abstract: There are provided a method for operating multi-node data stores. The method performs storing a data table in a first computing node and stores an index table in a second computing node. The index table provides keys used for accessing data in the first computing node and other multi-node data stores. The method performs operations that update or read the data table accessed from the first computing node and the index table accessed from the second computing node. The operations optimizes between latency in updating or reading the data table and the index table and data consistency maintained between data entries in the data table and data entries pointed by indices in the index table.Type: ApplicationFiled: December 21, 2015Publication date: June 30, 2016Inventors: Wei Tan, Sandeep Tata