Google Inc.

Microsoft Technology Licensing, LLC

Alta Vista Company

Hewlett-packard Development Company, L.p.

Other

Marc Alexander Najork

Mike Bendersky

Clark Allan Heydon

Michael Bendersky

Mingyang Zhang

Xuanhui Wang

Chandramohan Amyangot Thekkath

Sandeep Tata

Qi Zhao

Rina Panigrahy

Cheng Li

Donald Metzler

James Bradley Wendt

Navneet Potti

Liu Yang

Spurthi Amba Hombaiah

Bodhisattwa Prasad Majumder

Timothy Youngjin Sohn

Sujith Ravi

Vladimir Ofitserov

Rajesh Shenoy

Thomas Nelson

Peter Shao-sen Young

Tao Chen

Monika Hildegard Henzinger

Sreenivas Gollapudi

Mark Steven Manasse

Vanja Josifovski

Jie Yang

Janet Lynn Wiener

Luis Garcia Pueyo

Alexander Johannes Smola

MyLinh Yang

Hui Tan

Amitabh Saikia

Marc-Allen Cartright

Timothy Lewis Harrington

Michael David Mitzenmacher

Alexandros Ntoulas

Dennis Craig Fetterly

Amr Ahmed

Jing Xie

Karthik Raman

Atish Das Sarma

Remy Burger

Brandon Vargo

Jiecao Chen

Dor Naveh

Vlad Panait

Abbas Kazerouni

Weize Kong

Te-Lin Wu

Mike Colagrosso

Shuguang Han

Sergey Novikov

Przemysław Gajda

Weinan Zhang

Timothy Harrington

Systems and methods for document processing that can process and understand the layout, text size, text style, and multimedia of a document can generate more accurate and informed document representations. The layout of a document paired with text size and style can indicate what portions of a document are possibly more important, and the understanding of that importance can help with understanding of the document. Systems and methods utilizing a hierarchical framework that processes the block-level and the document-level of a document can capitalize on these indicators to generate a better document representation.

Layout-aware multimodal pretraining for multimodal document understanding

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.

System for information extraction from form-like documents

Implementations relate to training a model that can be used to process values for defined features, where the values are specific to a user account, to generate a predicted user measure that reflects both popularity and quality of the user account. The model is trained based on losses that are each generated as a function of both a corresponding generated popularity measure and a corresponding generated quality measure of a corresponding training instance. Accordingly, the model can be trained to generate, based on values for a given user account, a single measure that reflects both quality and popularity of the given user account. Implementations are additionally or alternatively directed to utilizing such predicted user measures to restrict provisioning of content items that are from user accounts having respective predicted user measures that fail to satisfy a threshold.

Training and/or utilizing a model for predicting measures reflecting both quality and popularity of content

Systems and methods of the present disclosure are directed to a method for predicting semantic similarity between documents. The method can include obtaining a first document and a second document. The method can include parsing the first document into a plurality of first textual blocks and the second document into a plurality of second textual blocks. The method can include processing each of the plurality of first textual blocks and the second textual blocks with a machine-learned semantic document encoding model to obtain a first document encoding and a second document encoding. The method can include determining a similarity metric descriptive of a semantic similarity between the first document and the second document based on the first document encoding and the second document encoding.

Systems and methods for machine-learned prediction of semantic similarity between documents

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for performing a machine learning task on a tuple of respective input sequences to generate an output. In one aspect, one of the systems includes a neural network comprising a plurality of encoder neural networks and a head neural network, each encoder neural network configured to: receive a respective input sequence from the tuple; process the respective input sequence using one or more encoder network layers to generate an encoded representation comprising a sequence of tokens; and process each of some or all of the tokens in the sequence of tokens using a projection layer to generate a lower-dimensional representation, and the head neural network configured to: receive lower-dimensional representations of a respective proper subset of the sequence of tokens generated by the encoder neural network; and process the lower-dimensional representations to generate the output.

Processing large-scale textual inputs using neural networks

Techniques of generating recrawl policies for commercial offer pages include generating a multiple strategy approach using a number of different strategies. In some implementations, each strategy is an arm of a K-armed adversarial bandits algorithm with reinforcement learning. Moreover, in some implementations, the multiple strategy approach also uses a machine learning algorithm to estimate parameters such as a click rate, impression rate, and likelihood of price change, i.e., change rate, which was assumed known in the conventional approaches.

Adversarial bandits policy for crawling highly dynamic content

Techniques are described herein for enabling more computationally efficient organization of files within a cloud storage system. A method includes: receiving information identifying a document and a set of folders; for each folder in the set of folders, using a trained model to predict a similarity measure between the folder and the document; for each folder in the set of folders, determining a score for the folder based on the predicted similarity measure for the folder; selecting a candidate folder from the set of folders using the scores of the folders within the set of folders; and providing, on a user interface, a selectable option to associate the document with the candidate folder.

Automatic file organization within a cloud storage system

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.

Systems and methods for active learning

Methods, systems, apparatus, including computer programs encoded on computer storage medium, to facilitate identification of additional trigger-terms for a structured information card. In one aspect, the method includes actions of accessing data associated with a template for presenting structured information, wherein the accessed data references (i) a label term and (ii) a value. Other actions may include obtaining a candidate label term, identifying one or more entities that are associated with the label term, identifying one or more of the entities that are associated with the candidate label term, and for each particular entity of the one or more entities that are associated with the candidate label term, associating, with the candidate label term, (i) a label term that is associated with the particular entity, and (ii) the value associated with the label term.

Search and retrieval of structured information cards

Methods and apparatus related to using document feature(s) of a document that is responsive to a query, and optionally query feature(s) of the query, to determine a presentation characteristic for presenting a search result that corresponds to the document. In some implementations, measures associated with the document feature(s) and/or query feature(s) may be used to determine the presentation characteristic. The measures may be based on past interactions, by corresponding users, with other documents that share one or more of the document features with the document, where a plurality of the other documents are different from the document (and optionally each different from one another). In some implementations, the document and/or the other documents include, or are restricted to, documents that are access restricted.

Ranking search result documents

Growing community of inventors

Palo Alto, CA, United States of America

Marc Alexander Najork