AT&T Intellectual Property I, LP

Facebook, Inc.

Nuance Communications, Inc.

Meta Platforms, Inc.

Linkedin Corporation

Sumit Chopra

Patrick Guy Haffner

Suhrid Balakrishnan

Jason Edward Weston

Eric Martin Zavesky

Ann K Syrdal

Taniya Mishra

Ilija Zeljkovic

David Lee Applegate

Simon Urbanek

Srinivas Bangalore

Dimitrios Dimitriadis

Keith Adams

Lubomir Dimitrov Bourdev

Balamanohar Paluri

Robert D Fergus

Alexander Matthew Rush

Du Le Hong Tran

Piotr Mirowski

Michael Auli

Antoine Bordes

Marc Aurelio Ranzato

Wojciech Zaremba

Disclosed herein are systems, methods, and non-transitory computer-readable storage media for performing speaker verification. A system configured to practice the method receives a request to verify a speaker, generates a text challenge that is unique to the request, and, in response to the request, prompts the speaker to utter the text challenge. Then the system records a dynamic image feature of the speaker as the speaker utters the text challenge, and performs speaker verification based on the dynamic image feature and the text challenge. Recording the dynamic image feature of the speaker can include recording video of the speaker while speaking the text challenge. The dynamic feature can include a movement pattern of head, lips, mouth, eyes, and/or eyebrows of the speaker. The dynamic image feature can relate to phonetic content of the speaker speaking the challenge, speech prosody, and the speaker's facial expression responding to content of the challenge.

System and method for dynamic facial features for speaker recognition

In one embodiment, a method includes receiving, from a client system, a text input comprising one or more n-grams, determining, using a deep-learning model, a vector representation of the text input based on the one or more n-grams, determining an embedding of the vector representation of the text input in a d-dimensional embedding space, identifying one or more labels based on, for each of the one or more labels, a respective similarity of an embedding of a vector representation of the label in the embedding space to the embedding of the vector representation of the text input, and sending, to the client system in response to the received text input, instructions for presenting a user interface comprising one or more of the identified labels as suggested labels.

Predicting labels using a deep-learning model

In one embodiment, a sequence of input words is received. Each of the input words is encoded as an indicator vector, wherein a sequence of the indicator vectors captures features of the sequence of input words. The sequence of the indicator vectors is then mapped to a distribution of a contextual probability of a first output word in a sequence of output words. For each subsequent output word, the sequence of the indicator vectors is encoded with a context, wherein the context comprises a previously mapped contextual probability distribution of a fixed window of previous output words; and the encoded sequence of the indicator vectors and the context is mapped to the distribution of the contextual probability of the subsequent output word. Finally, a condensed summary is generated using a decoder by maximizing the contextual probability of each of the output words.

Abstractive sentence summarization

Embodiments are disclosed for providing a machine-generated response (e.g., answer) to an input (e.g., question) based on long-term memory information. A method according to some embodiments include receiving an input; converting the input into an input feature vector in an internal feature representation space; updating a memory data structure by incorporating the input feature vector into the memory data structure; generating an output feature vector in the internal feature representation space, based on the updated memory data structure and the input feature vector; converting the output feature vector into an output object; and providing an output based on the output object as a response to the input.

Generating responses using memory networks

In one embodiment, a method includes accessing a first set of entities, with which a user has interacted, and a second set of entities in a social-networking system. A first set of vector representations of the first set of entities are determined using a deep-learning model. A target entity is selected from the first set of entities, and the vector representation of the target entity is removed from the first set. The remaining vector representations in the first set are combined to determine a vector representation of the user. A second set of vector representations of the second set of entities are determined using the deep-learning model. Similarity scores are computed between the user and each of the target entity and the entities in the second set of entities. Vector representations of entities in the second set of entities are updated based on the similarity scores using the deep-learning model.

Identifying entities using a deep-learning model

A system for training a model to predict a sequence (e.g. a sequence of words) given a context is disclosed. A model can be trained to make these predictions using a combination of individual predictions compared to base truth and sequences of predictions based on previous predictions, where the resulting sequence is compared to the base truth sequence. In particular, the model can initially use the individual predictions to train the model. The model can then be further trained over the training data in multiple iterations, where each iteration includes two processes for each training element. In the first process, an initial part of the sequence is predicted, and the model and model parameters are updated after each prediction. In the second process, the entire remaining amount of the sequence is predicted and compared to the corresponding training sequence to adjust model parameters to encourage or discourage each prediction.

Training sequence natural language processing engines

In one embodiment, a method includes receiving text query that includes n-grams. A vector representation of each n-gram is determined using a deep-learning model. A nonlinear combination of the vector representations of the n-grams is determined, and an embedding of the text query is determined based on the nonlinear combination. The embedding of the text query corresponds to a point in an embedding space, and the embedding space includes a plurality of points corresponding to a plurality of label embeddings. Each label embedding is based on a vector representation of a respective label determined using the deep-learning model. Label embeddings are identified as being relevant to the text query by applying a search algorithm to the embedding space. Points corresponding to the identified label embeddings are within a threshold distance of the point corresponding to the embedding of the text query in the embedding space.

A system that incorporates teachings of the subject disclosure may include, for example, determining identified impressions that are detected from consumption data collected from a group of media processors where the identified impressions represent viewing of selected content and where the consumption data indicates channel tuning events at the group of media processors including changing of channels, applying a ridge regression analysis to the identified impressions to determine a predicted number of target impressions per advertisement slot, and generating a media plan based on a ratio of an advertisement slot cost to the predicted number of target impressions per advertisement slot. Other embodiments are disclosed.

Method and apparatus for managing advertising

Systems, methods, and non-transitory computer-readable media can acquire video content for which video feature descriptors are to be determined. The video content can be processed based at least in part on a convolutional neural network including a set of two-dimensional convolutional layers and a set of three-dimensional convolutional layers. One or more outputs can be generated from the convolutional neural network. A plurality of video feature descriptors for the video content can be determined based at least in part on the one or more outputs from the convolutional neural network.

Systems and methods for determining video feature descriptors based on convolutional neural networks

Disclosed herein are systems, methods, and non-transitory computer-readable storage media for combining frame and segment level processing, via temporal pooling, for phonetic classification. A frame processor unit receives an input and extracts the time-dependent features from the input. A plurality of pooling interface units generates a plurality of feature vectors based on pooling the time-dependent features and selecting a plurality of time-dependent features according to a plurality of selection strategies. Next, a plurality of segmental classification units generates scores for the feature vectors. Each segmental classification unit (SCU) can be dedicated to a specific pooling interface unit (PIU) to form a PIU-SCU combination. Multiple PIU-SCU combinations can be further combined to form an ensemble of combinations, and the ensemble can be diversified by varying the pooling operations used by the PIU-SCU combinations. Based on the scores, the plurality of segmental classification units selects a class label and returns a result.

System and method for combining frame and segment level processing, via temporal pooling, for phonetic classification

Disclosed herein are systems, methods, and non-transitory computer-readable storage media for learning latent representations for natural language tasks. A system configured to practice the method analyzes, for a first natural language processing task, a first natural language corpus to generate a latent representation for words in the first corpus. Then the system analyzes, for a second natural language processing task, a second natural language corpus having a target word, and predicts a label for the target word based on the latent representation. In one variation, the target word is one or more word such as a rare word and/or a word not encountered in the first natural language corpus. The system can optionally assigning the label to the target word. The system can operate according to a connectionist model that includes a learnable linear mapping that maps each word in the first corpus to a low dimensional latent space.

Growing community of inventors

Jersey City, NJ, United States of America

Sumit Chopra