Intuit, Inc.

Qualcomm Incorporated

Snap Inc.

Vitor R Carvalho

Babak Forutanpour

Igor A Podgorny

Sparsh Gupta

Ning Xu

Jia Li

Ning Zhang

Shivakumar Balasubramanyam

Xiaoyu Wang

Matthew Cannon

Peter Ouyang

Faraz Sharafi

Janani Kalyanam

Leah Zhao

Edward Harrison Teague

Carlos Manuel Puig

Shankar Sadasivam

Bennett Michael King

Jill Sayuri Iwasaki

Bhuwan Dhingra

Siddika Parlak Polatkan

Michael R Cowgill

Geoffrey C Wenger

Anusha Kamath

Bennett M King

Systems, methods, devices, media, and computer readable instructions are described for local image tagging in a resource constrained environment. One embodiment involves processing image data using a deep convolutional neural network (DCNN) comprising at least a first subgraph and a second subgraph, the first subgraph comprising at least a first layer and a second layer, processing, the image data using at least the first layer of the first subgraph to generate first intermediate output data; processing, by the mobile device, the first intermediate output data using at least the second layer of the first subgraph to generate first subgraph output data, and in response to a determination that each layer reliant on the first intermediate data have completed processing, deleting the first intermediate data from the mobile device. Additional embodiments involve convolving entire pixel resolutions of the image data against kernels in different layers if the DCNN.

Content tagging

Aspects of the present disclosure provide techniques for FAQ retrieval. Embodiments include receiving, via a user interface of a computing application, a query related to a subject. Embodiments include generating a first multi-dimensional representation of the query. Embodiments include obtaining a plurality of question and answer pairs related to the subject and, for a given question and answer pair comprising a given question and a given answer, generating a second multi-dimensional representation of the given question and a third multi-dimensional representation of the given answer. Embodiments include providing input to a model based on the first multi-dimensional representation, the second multi-dimensional representation, and the third multi-dimensional representation and determining a match score for the query and the given question and answer pair based on an output of the model. Embodiments include providing, via the user interface of the computing application, the question and answer pair based on the match score.

Retrieval of frequency asked questions using attentive matching

A method for computer estimations based on statistical tree structures involves obtaining a statistical tree structure for reference elements. The statistical tree structure includes leaf nodes segmenting a statistic for a data label according to data features in the reference elements, and intermediate nodes connecting a first node to the leaf nodes. Each of the first node and the intermediate nodes provide a branching based on one of the data features. The method further includes obtaining target data, including values for the data features, and a value for the data label. The method also includes selecting the first node, associated with a first data feature, traversing the statistical tree structure to a leaf node by matching the values of the data features to the branching of the intermediate nodes, and assessing the value for the data label in the target data based on the statistic associated with the leaf node.

Computer estimations based on statistical tree structures

A method for facilitating user support using multimodal information involves obtaining an interaction between a user and a support agent, generating a question embedding from the interaction, obtaining a clickstream associated with the interaction, and generating a clickstream embedding from the clickstream. The question embedding and the clickstream embedding form a shared latent space representation. The method further involves decoding a problem summary from the shared latent space representation and providing the problem summary to the support agent.

Method and system for facilitating user support using multimodal information

A computer-implemented method is provided to perform text classification with a neural network system. The method includes providing a computing device to receive input datasets including user input question text and feed the datasets to the neural network system. The neural network system includes one or more neural networks configured to extract and concatenate character-based features, word-based features from the question datasets and clickstream embeddings of clickstream data to form a representation vector indicative of the question text and user behavior. A representation vector is fed into fully connected layers of a feed-forward network. The feed-forward network is configured to predict a first class and a second class associated with respective user input questions based on the representation vector.

Neural network system for text classification

A method for text classification involves generating, using a bilingual embedding model, source language embeddings for source language documents; obtaining source language document labels of the source language documents; and training a source language classifier model and a label embedding network, executing on a computing system, using the source language embeddings and the source language document labels. The method further involves generating pseudo-labels for unlabeled target language documents, by: generating, using the bilingual embedding model, target language embeddings for the unlabeled target language documents, and applying the source language classifier model and the label embedding network to the target language embeddings to obtain the pseudo-labels for the unlabeled target language documents. In addition, the method involves training a target language classifier model executing on the computing system using the target language embeddings and the pseudo labels.

Cross-language models based on transfer learning

Disclosed is an apparatus and method for classifying a motion state of a mobile device comprising: determining a first motion state associated with a highest probability value and with a first confidence level greater than a first threshold; entering the first motion state; while the first motion state is active, determining a second motion state associated with a highest probability value and with a second confidence level greater than the first threshold, the second motion state being different from the first motion state; determining whether the second motion state is to be entered; and in response to determining that the second motion state is to be entered, entering the second motion state.

Smartphone motion classifier

Detect duplicated questions using reverse gradient adversarial domain adaptation includes applying a general network to multiple general question pairs to obtain a first set of losses. A target domain network is applied to multiple domain specific network pairs to obtain a second set of losses. Further, a domain distinguishing network is applied to a set of domain specific questions and a set of general questions to obtain a third set of losses. A set of accumulated gradients is calculated from the first set of losses, the second set of losses, and the third set of losses. Multiple features are updated according to the set of accumulated gradients to train the target domain network.

Detecting duplicated questions using reverse gradient adversarial domain adaptation

A method for decoding a natural language user query involves obtaining the user query submitted by a user, segmenting the user query into words, generating a character embedding for each of the words, and generating a word embedding for each of the words. The method further involves obtaining a clickstream from tracked clicks of the user, generating a clickstream embedding from the clickstream, and for each of the words, generating a unified feature representation based on the character embedding and the word embedding for each of the words, and the clickstream embedding. The method also involves decoding the unified feature representations to obtain a decoded user query, and processing the user query using the decoded user query.

Method and system for decoding user intent from natural language queries

Systems and methods for content tagging

An apparatus, a method, and a computer program product for gesture recognition. The apparatus classifies a gesture based on a movement of a body part as detected by a primary sensor. The apparatus determine a reliability level of a secondary sensor and obtains corroborating information associated with the movement of the body part using the secondary sensor when the reliability level satisfies a criterion. The apparatus then confirms or negates the classification of the gesture based on the corroborating information. The secondary sensor may be a sensor already known to the apparatus, i.e., the sensor is currently being worn by the user, or it may be a sensor that is worn by a user at a later time. In the latter case, the apparatus detects for the presence of a new sensor, determines the gesture recognition capabilities of the new sensor and integrates the new sensor into the gesture recognition process.

Classification of gesture detection systems through use of known and yet to be worn sensors

Various arrangements for recognizing a gesture are presented. User input may be received that causes a gesture classification context to be applied from a plurality of gesture classification contexts. This gesture classification context may be applied, such as to a gesture analysis engine. After applying the gesture classification context, data indicative of a gesture performed by a user may be received. The gesture may be identified in accordance with the applied gesture classification context.

Context-sensitive gesture classification

A system, a method, and a computer program product for managing information for an interface device are provided. The system detects for one of a present physical encounter between a user of the interface device and a person, and a non-physical encounter between the user and a person. The system determines if a detected present physical encounter is an initial encounter or a subsequent encounter, and adds content associated with the person to a database of previously encountered persons when the present physical encounter is an initial encounter. When the present physical encounter is a subsequent encounter or a present non-physical encounter is detected, the system determines if the person is known by the user, and presents information to the interface device corresponding to the person when the person is not known by the user.

Growing community of inventors

San Diego, CA, United States of America

Vitor R Carvalho