Tubi Limited

Adobe, Inc.

Jaya Kawale

Amir Mazaheri

Chang She

Marios Assiotis

Hailong Geng

Yu Tao Zhao

Yoshihiro Tanaka

Haofei Wang

Mark Alden Rotblat

Joseph Gallagher

Khaldun Matter Ahmad AlDarabsah

Chiyu Zhong

Branislav Kveton

Hung Hai Bui

Nikolaos Vlassis

Suvash Sedhain

Sheng Li

Dennis Paul Yost

Kevin Edward Corcoran

John Trenkle

Matthew Adam Elliott

Anthony Albert Truyoo-Broque

Systems and methods for programmatic generation of training data, including: a training data generation engine configured to: identify an image asset corresponding to an entity; identify a training video; select a consecutive subset of frames of the training video based on a procedure for ranking frames on their candidacy for overlaying content; for at least one frame of the subset of frames: perform an augmentation technique on the identified logo image to generate an augmented image asset; overlay at least one variation of the image asset, including the augmented image asset, onto each of the subset of frames to generate a set of overlayed frames; and generate an augmented version of the training video including the overlayed frames; and a model training engine configured to: train an artificial intelligence model for entity detection using the augmented version of the training video.

Training data generation for advanced frequency management

Systems and methods for frequency management, including: an online media service configured to receive a request for a media item, the request comprising a recipient identifier of a recipient, and identify a set of candidate media items ranked based at least partially on relevance to the recipient; and a frequency management service configured to: (i) identify a quantity of impressions associated with a first candidate media item of the set of candidate media items and the recipient identifier over a preceding duration of time, (ii) identify a maximum frequency threshold, (iii) determine, based on the quantity of impressions, that the maximum frequency threshold is exceeded, (iv) exclude the first candidate media item from a result set based on the frequency threshold being exceeded, and (v) provide the result set comprising an identifier of a second candidate media item in response to the request.

Machine learning techniques for advanced frequency management

A system and method for multimedia scene break detection, including: a computer processor, a scene break detection service executing on the computer processor and comprising functionality to: (i) receive a request for scene break detection on the media item, (ii) perform audio break detection on an audio component of the media item to obtain a set of audio break timestamps, (iii) identify a set of video break timestamps, each corresponding to at least one frame of a video component of the media item, (iv) identify a set of candidate scene break timestamps based on the audio break timestamps and the video break timestamps, (v) execute a computer vision scoring model for each candidate scene break timestamp, and (vi) select, based at least on a score of each of the set of candidate scene break timestamps, a final set of scene break timestamps for performing a media action.

Multimedia scene break detection

Systems and methods for entity detection using artificial intelligence, including: a deep learning model service configured to: select and analyze a set of frames from a media item to determine a set of candidate brand-probability pairs; a voting engine configured to: determining that a first brand-probability pair of a set of candidate brand-probability pairs based on at least one obtained hyperparameter value does not meet a threshold for determining whether candidate brand-probability pairs are to be included in a result set; excluding the first brand-probability pair from the result set based on the determination; sorting the result set; and selecting at least one final brand-probability pair from the result set; and an offline transcoding service configured to: store the final brand-probability pair in a repository with a relation to an identifier of the media item.

Model serving for advanced frequency management

Systems and methods for frequency management, including: an online media service configured to: receive a request for a media item, the request including a recipient identifier; identify a set of candidate media items ranked by a set of matching criteria; a frequency management service configured to: perform a query against a lookup service, where the query includes (i) an entity identifier of at least one candidate media item of the set of candidate media items, and (ii) the recipient identifier; receive a response from the lookup service including a quantity of impressions associated with the entity identifier and the recipient identifier; identify a predefined frequency threshold; determine that the frequency threshold is exceeded and exclude the at least one candidate media item from a result set based on the determination; and provide the result set including an identifier of at least one other candidate media item.

Techniques disclosed herein provide more efficient and more relevant item recommendations to users in large-scale environments in which only positive interest information is known. The techniques use a rank-constrained formulation that generalizes relationships based on known user interests in items and/or use a randomized singular value decomposition (SVD) approximation technique to solve the formulation to identify items of interest to users in an efficiently, scalable manner.

Item recommendation techniques

Particle Thompson Sampling for online matrix factorization recommendation is described. In one or more implementations, a recommendation system provides a recommendation of an item to a user using Thompson Sampling. The recommendation system then receives a rating of the item from the user. Unlike conventional solutions which only update the user latent features, the recommendation system updates both user latent features and item latent features in a matrix factorization model based on the rating of the item. The updating is performed in real time which enables the recommendation system to quickly adapt to the user ratings to provide new recommendations. In one or more implementations, to update the user latent features and the item latent features in the matrix factorization model, the recommendation system utilizes a Rao-Blackwellized particle filter for online matrix factorization.

Particle thompson sampling for online matrix factorization recommendation

A deep collaborative filtering (DCF) approach is employed in a recommender system to provide item recommendations to users. The DCF approach combines deep learning models with matrix factorization based collaborative filtering. To provide item recommendations, a user-item rating matrix, user side information, and item side information are provided as input to a recommender system. The recommender system learns user latent factors and item latent factors by jointly: (1) decomposing the user-item rating matrix to extract latent factors, and (2) extracting latent factors from hidden layers of deep learning models using the user side information and item side information. The learned user latent factors and item latent factors are used to predict item ratings for missing ratings in the user-item rating matrix. The predicted item ratings are then used to select item recommendations for a given user, which are then communicated to a user device of the user.

Growing community of inventors

San Jose, CA, United States of America

Jaya Kawale