Google Inc.

Yahoo! Inc.

Sujith Ravi

Andrew Stephen Tomkins

Robin Dua

John Patrick McGregor, Jr

Zornitsa Kozareva

Bo Pang

Ariel Fuxman

Amit Fulay

Ryan H Cassidy

Vivek Ramavajjala

Sergey Igorevich Nazarov

Shanmugasundaram Ravikumar

Marc Alexander Najork

Timothy Youngjin Sohn

Thomas J Duerig

Tristan Harris

Michael Bendersky

Can Sar

Angelo DiNardi

Mingyang Zhang

Zhen Li

Futang Peng

Xuanhui Wang

Da-Cheng Juan

Aleksei Timofeev

Chun-Ta Lu

Thomas Nelson

Peter Shao-sen Young

Qiming Diao

Yi-ting Chen

Yaxi Gao

Nathan Dickerson Beach

Gaurav Menghani

Chinnadhurai Sankar

Thomas Matthew Rudick

Mirko Ranieri

Prabhu Kaliamoorthi

Dana Movshovitz-Attias

Gaurav Nemade

Yicheng Fan

Jeongwoo Ko

Dora Demszky

The technology relates to methods for detecting and classifying emotions in textual communication, and using this information to suggest graphical indicia such as emoji, stickers or GIFs to a user. Two main types of models are fully supervised models and few-shot models. In addition to fully supervised and few-shot models, other types of models focusing on the back-end (server) side or client (on-device) side may also be employed. Server-side models are larger-scale models that can enable higher degrees of accuracy, such as for use cases where models can be hosted on cloud servers where computational and storage resources are relatively abundant. On-device models are smaller-scale models, which enable use on resource-constrained devices such as mobile phones, smart watches or other wearables (e.g., head mounted displays), in-home devices, embedded devices, etc.

Methods for emotion classification in text

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training an image embedding model. In one aspect, a method comprises: obtaining training data comprising a plurality of training examples, wherein each training example comprises: an image pair comprising a first image and a second image; and selection data indicating one or more of: (i) a co-click rate of the image pair, and (ii) a similar-image click rate of the image pair; and using the training data to train an image embedding model having a plurality of image embedding model parameters.

Training image and text embedding models

Implementations are directed to facilitating user device and/or agent device actions during a communication session. An interactive communications system provides outputs, as outlined below, that are tailored to enhance the functionality of the communication session, reduce the number of dialog 'turns' of the communications session and/or the number of user inputs to devices involved in the session, and/or otherwise mitigate consumption of network and/or hardware resources during the communication session. In various implementations, the communication session involves user device(s) of a user, agent device(s) of an agent, and the interactive communications system. The interactive communications system can analyze various communications from the user device(s) and/or agent device(s) during a communication session in which the user (via the user device(s)) directs various communications to the agent, and in which the agent (via the agent device(s)) optionally directs various communications to the user. The interactive communications system provides action performance element(s) and/or other output(s) that are each specific to a corresponding current intent and corresponding current action of the communication session.

Facilitating user device and/or agent device actions during a communication session

The present disclosure provides projection neural networks and example applications thereof. In particular, the present disclosure provides a number of different architectures for projection neural networks, including two example architectures which can be referred to as: Self-Governing Neural Networks (SGNNs) and Projection Sequence Networks (ProSeqoNets). Each projection neural network can include one or more projection layers that project an input into a different space. For example, each projection layer can use a set of projection functions to project the input into a bit-space, thereby greatly reducing the dimensionality of the input and enabling computation with lower resource usage. As such, the projection neural networks provided herein are highly useful for on-device inference in resource-constrained devices. For example, the provided SGNN and ProSeqoNet architectures are particularly beneficial for on-device inference such as, for example, solving natural language understanding tasks on-device.

On-device projection neural networks for natural language understanding

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for a projection neural network. In one aspect, a projection neural network is configured to receive a projection network input and to generate a projection network output from the projection network input. The projection neural network includes a sequence of one or more projection layers. Each projection layer has multiple projection layer parameters, and is configured to receive a layer input, apply multiple projection layer functions to the layer input, and generate a layer output by applying the projection layer parameters for the projection layer to the projection function outputs.

Projection neural networks

Systems and methods are provided to pre-train projection networks for use as transferable natural language representation generators. In particular, example pre-training schemes described herein enable learning of transferable deep neural projection representations over randomized locality sensitive hashing (LSH) projections, thereby surmounting the need to store any embedding matrices because the projections can be dynamically computed at inference time.

Pre-trained projection networks for transferable natural language representations

The present disclosure provides an application development platform and associated software development kits ('SDKs') that provide comprehensive services for generation, deployment, and management of machine-learned models used by computer applications such as, for example, mobile applications executed by a mobile computing device. In particular, the application development platform and SDKs can provide or otherwise leverage a unified, cross-platform application programming interface ('API') that enables access to all of the different machine learning services needed for full machine learning functionality within the application. In such fashion, developers can have access to a single SDK for all machine learning services.

Application development platform and software development kits that provide comprehensive machine learning services

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

A method includes identifying images associated with a user, where the image is identified as at least one of captured by a user device associated with the user, stored on the user device associated with the user, and stored in cloud storage associated with the user. The method also includes for each of the images, determining one or more labels, wherein the one or more labels are based on at least one of metadata and a primary annotation. The method also includes generating a mapping of the one or more labels to one or more confidence scores, wherein the one or more confidence scores indicate an extent to which the one or more labels apply to corresponding images. The method also includes interacting with the user to obtain identifying information that is used to categorize one or more of the images.

Organizing images associated with a user

On-device neural networks for natural language understanding

Implementations relate to automatic response suggestions based on images received in messaging applications. In some implementations, a computer-executed method includes detecting a first image included within a first message received at a second device over a communication network from a first device of a first user, and programmatically analyzing the first image to extract a first image content. The method includes retrieving a first semantic concept associated with the first image content, programmatically generating a suggested response to the first message based on the first semantic concept, and transmitting instructions causing rendering of the suggested response in the messaging application as a suggestion to a second user of the second device.

Automatic response suggestions based on images received in messaging applications

A computing device may receive a communication sent from an external computing device. At least one processor of the computing device may determine, using an on-device machine-trained model and based at least in part on the communication, one or more candidate responses to the communication. The at least one processor may receive an indication of a user input that selects a candidate response from the one or more candidate responses. Responsive to receiving the indication of the user input that selects the candidate response, the at least one processor may send the candidate response to the external computing device.

Smart replies using an on-device model

Growing community of inventors

Santa Clara, CA, United States of America

Sujith Ravi