Fuji Xerox Co., Ltd.

Fujifilm Business Innovation Corp.

Yin-Ying Chen

Francine R Chen

Dhiraj Joshi

David Ayman Shamma

Lyndon Kennedy

Jian Zhao

Bor-Chun Chen

Scott Carter

Qiong Liu

Matthew L Cooper

Yanxia Zhang

Shigeyuki Sakaki

Hao Hu

Bokai Cao

Cheng Zhang

Mandy Barrett Korpusik

Jyun-Yu Jiang

Ray Yuan

Heike Adel

Philipp Daniel Harzig

Sanjeev Kumar Karn

A computer-implemented method of learning sensory media association includes receiving a first type of nontext input and a second type of nontext input; encoding and decoding the first type of nontext input using a first autoencoder having a first convolutional neural network, and the second type of nontext input using a second autoencoder having a second convolutional neural network; bridging first autoencoder representations and second autoencoder representations by a deep neural network that learns mappings between the first autoencoder representations associated with a first modality and the second autoencoder representations associated with a second modality; and based on the encoding, decoding, and the bridging, generating a first type of nontext output and a second type of nontext output based on the first type of nontext input or the second type of nontext input in either the first modality or the second modality.

System and method for learning sensory media association without using text labels

A method and system for classifying image features using a neural network is provided. The method includes training the neural network using triplet loss processes including receiving an anchor image, selecting a positive image and a negative image, generating a image embedding associated with each of the anchor image, the positive image, and the negative image, classifying image features extracted from the anchor image based on the image embedding of the anchor image, determining an image label location associated with the classified image features, extracting features associated with the determined image label location, and classifying the features associated with the determined image label location; and combining the multi-label loss with localized image classification loss and the triplet loss using a weighted loss sum.

System and method for identification and localization of images using triplet loss and predicted regions

A method of analyzing unstructured messages is provided. The method includes extracting a content feature from an embedding associated with each message from a pair of unstructured messages using a neural network, generating, based on extracted content features, a text similarity score between the pair of messages using a second neural network, combining the generated text similarity score with additional data associated with the messages to generate a total similarity score, and generating a message thread based on the generated total similarity score for the pair of messages selected from the plurality of unstructured messages.

System and method for detangling of interleaved conversations in communication platforms

A computer implemented method of grouping media objects is provided, as well as systems, interfaces and devices therefor. The method includes generating a group from the media objects based on a combination of a script of sequential events and an actor associated with one or more of the media objects in the script, segmenting the group into segments each including one or more of the media objects, based on clustering or classification, providing titling and captioning for the segments, and generating filter and annotation recommendations based on knowledge associations in the media objects, data, and the combination of the script and the actor, across the media objects of the group.

Media object grouping and classification for predictive enhancement

A method and system of targeting information for a user is provided. The method includes collecting one or more digital posts that identify a product and are associated with a digital account, identifying, by a neural network, a stage of a decision model associated with each of the one or more digital posts based on a feature representation generated for each digital post, and transmitting product-related information to the digital account, wherein the transmitting is based on the stage of the decision model identified for each of the one or more digital posts.

System and method for using deep learning to identify purchase stages from a microblog post

A method and device for displaying a plurality of content items may be shown. The method may include detecting at least one topic associated with each of the plurality of content items, generating a set of topics based on the at least one topic associated with each of the plurality of content items, the set of topics comprising a number of topics associated with the plurality of content items, reducing the number of topics in the set of topics, by combining one of the number of topic in response to a determination that at least one topic in the set of topics is redundant, and generating a visualization, the visualization including at least one content item from the plurality of content items; and the set of topics.

System and method for user-oriented topic selection and browsing

A method and system for determining a treatment order for a plurality of patient imaging records. The method includes extracting, by a trained neural network, image features from each of the plurality of imaging records, generating, by the trained neural network, a written report associated with each of the plurality of imaging records based on the extracted image features, wherein the trained neural network generates the written report based on a sentence annotation model that provides abnormality annotations on an individual sentence basis, determining, by the trained neural network, an abnormality score associated with each written report, and providing the written reports to a treating physician in a sorted order based on the abnormality score associated with each written report.

System and method for generating descriptions of abnormalities in medical images

A method and system for generating summaries of posts of interleaved text are provided. The method includes embedding, by a first neural network, each post through word-to-word encoding; embedding, by a second neural network, overall content of the plurality of posts through post-to-post encoding based on the word-to-word encoding of each post; generating, by at least a third neural network, a summary of the at least one thread through word-to-word decoding based on the overall content embedding of the plurality of posts; and displaying the summary of the at least one thread to a user.

System and method for generating abstractive summaries of interleaved texts

A method and system of associating textual summaries with data representative of media content is provided. The method may include receiving a plurality of textual summaries, each textual summary representative of an event, pairing, by a neural network, each received textual summary with each of a plurality of pieces of data, each piece of data representative of media content, to generate a plurality of text-data pairings; and associating a first selected textual summary with a first piece of data based on a similarity of content features extracted from each received textual summary to content features extracted from each piece of data in each of the plurality of text-data pairings.

System and method for associating textual summaries with content media

A method and system for creating a visual representation of data is provided. The method may include receiving a user input to be represented by a glyph, generating, by a neural network, the glyph based on a vector extracted based on the received user input, linking an aspect of the generated glyph with an attribute of the data to be visually represented; and displaying the visual representation of the data created using the generated glyph to represent that linked aspect of the data.

System and method for creating visual representation of data based on generated glyphs

A method of generating a summary of a media file that comprises a plurality of media segments is provided. The method includes calculating, by a neural network, respective importance scores for each of the media segments, based on content features associated with each of the media segments and a targeting approach, selecting a media segment from the media segments, based on the calculated importance scores, generating a caption for the selected media segment based on the content features associated with the selected media segment, and generating a summary of the media file based on the caption.

Method and system to generate targeted captions and summarize long, continuous media files

A method of generating a caption for a social media post is provided. The method may include receiving a social media post to be posted to a social media network; collecting reference data relevant to determining common activities occurring at a category of location associated with the social media post; inferring potential topics for captions from a topic inference model, based on the collected reference data associated with the social media post; setting parameters associated with a language model based on the inferred topic; and generating at least one caption for the social media post based on the identified language model, and the inferred topic.

Method and apparatus for using business-aware latent topics for image captioning in social media

Example implementations include a system and method of recognizing behavior of a user. In example implementations, a first post and at least one subsequent post indicative of a product and associated with a first social media account is obtained. A relevance probability is calculated for each of the obtained first post and the at least one subsequent post. The obtained first post and the at least one subsequent post are sequentially analyzed by a second neural network to determine output values relevant to probability of purchasing the product. A probability of purchasing the product is calculated based on the determined output values associated with each post and the calculated relevance probabilities. Product-related information is transmitted to the user associated with the obtained first post based on the determined probability of purchasing the product.

Behavior prediction on social media using neural networks

A method associates social media messages with venues. A social network graph includes users, messages from users, and venues. The venues include multiple primary venues and a no-venue. A link between a message and the no-venue node indicates that the message is not associated with a primary venue. Training feature vectors are constructed that measure connectedness between messages and venues. The process trains a classifier to estimate probabilities that messages are associated with venues. A new social media message is received, and the process constructs a feature vector using the same features as the training vectors, measuring connectedness between the new message and the no-venue. The classifier computes a probability that the new message is associated with the no-venue. When the probability exceeds a predefined threshold, the new message is not associated with any of the primary venues. Otherwise, the new message is associated with one of the primary venues.

Venue link detection for social media messages

Systems and methods disclosed herein associate images with business venues. An example method includes: receiving a first image and retrieving textual reviews and stored images that are associated with one or more candidate business venues. The method further includes: detecting, using trained visual detectors, a plurality of business-aware concepts in the first image and assessing likelihood that detected business-aware concepts are in the first image. The method additionally includes: (i) generating a first representation of the first image based on the likelihoods and one or more term vectors for high-scoring concepts and (ii) receiving second representations of each candidate based on the retrieved textual reviews and stored images. In accordance with determining that the first representation is most similar to a respective second representation of a first candidate, the method includes: (i) associating the first image with the first candidate and (ii) providing an indication of the association.

Growing community of inventors

Sunnyvale, CA, United States of America

Yin-Ying Chen