Sony Interactive Entertainment Inc.

Saket Kumar

Steven Osman

Katrine Chow

Lakshmish Kaushik

Xavier Menendez-Pidal

Yuichiro Nakamura

Masanori Omote

Zhenhao Ge

Komath Naveen Kumar

Techniques are described for training neural networks on variable length datasets. The numeric representation of the length of each training sample is randomly perturbed to yield a pseudo-length, and the samples sorted by pseudo-length to achieve lower zero padding rate (ZPR) than completely randomized batching (thus saving computation time) yet higher randomness than strictly sorted batching (thus achieving better model performance than strictly sorted batching).

Semi-sorted batching with variable length input for efficient training

Text and speech from a computer simulation are processed by a machine learning engine to animate the face of a computer avatar.

Facial animation control by automatic generation of facial action units using text and speech

Methods and systems are provided for generating a video stream for followers of an influencer during a session. The method includes capturing video images of an environment of the influencer, wherein the environment includes the influencer viewing an online content channel. The method includes detecting surface parameters in the environment and processing the surface parameters to identify three-dimensional features associated with the surface parameter. The method includes identifying a follower that is viewing the influencer. The method includes generating an overlay image for at least one of the surface parameters, the overlay image is configured for dynamic contouring to the identified three-dimensional features during the session. The overlay image is selected based on a profile of the follower so that the video stream received by the follower when viewing the influencer is customized based on the profile of the follower.

Influencer stream customization for follower viewers

A device provides audible output pertaining to audio video (AV) content such as a video game based on a section of the AV content that is being presented.

Systems and methods to provide audible output based on section of content being presented

Methods and systems for improving engagement metrics of a spectator include identifying a group of spectators watching game play of a video game and generating an aggregate group profile for the group. Engagement metrics for the group are analyzed to identify engagement level of the group toward the game play of the player. One or more suggestions are provided to adjust game play of the video game so as to improve engagement level of the group toward the game play of the video game.

Methods and systems for improving spectator engagement in a video game

Methods and systems are provided for recommending content channels for an influencer. The method includes identifying a session of the influencer. The session includes a current content channel being viewed by the influencer and made available for streaming to one or more followers of the influencer. The current content channel is one of a plurality of content channels made available for viewing by the influencer. The method includes accessing, during the session, content data from the plurality of content channels. The content data is associated with gameplay of one or more players engaged in gameplay. The method includes accessing profile data of the followers of the influencer. The profile data includes content preferences of said followers. The method includes predicting engagement metrics for said followers of the influencer in relation to said plurality of content channels. The engagement metrics are updated in substantial real time based on change occurring with the content data in plurality of content channels. The method includes generating an influencer interface for presenting said predicted engagement metrics. The engagement metrics are configured to provide indicators of when engagement metrics for the current content channel indicate a predicted decrease relative to a predicted increase when switching to a different content channel from among the plurality of content channels available during the session.

Influencer tools for stream curation based on follower information

A method is provided, including the following operations: analyzing gameplay data and spectator data from previous sessions of a video game, wherein the analyzing is configured to correlate a spectator preference to a gameplay condition of the video game; using the correlated spectator preference to generate an in-game recommendation for the gameplay condition; identifying the gameplay condition occurring in a current session of the video game; responsive to identifying the gameplay condition occurring in the current session, then presenting the in-game recommendations to a player of the current session of the video game.

Growing community of inventors

San Mateo, CA, United States of America

Saket Kumar