USPTO Patent Grant US12586353B2: Unsupervised Learning of Object Representations

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published March 24th, 2026

Detected March 25th, 2026

Summary

The USPTO has granted patent US12586353B2 for an unsupervised learning system for object representation from video sequences using attention over space and time. The patent, assigned to GDM Holding LLC, details a neural network system for video generation and object latent variable determination.

View original document View source feed page

What changed

The United States Patent and Trademark Office (USPTO) has granted patent US12586353B2, titled 'Unsupervised learning of object representations from video sequences using attention over space and time.' This patent, assigned to GDM Holding LLC, covers a computer-implemented video generation neural network system designed to determine object latent variables and generate image frames. The system utilizes attention mechanisms over space and time for unsupervised learning of object representations from video sequences.

This patent grant is primarily an intellectual property matter and does not impose direct compliance obligations on regulated entities. However, it signifies innovation in the field of AI and video generation, potentially impacting companies involved in developing or utilizing such technologies. Companies operating in AI research, computer vision, and video generation should be aware of this patent as it may relate to their intellectual property landscape and potential licensing requirements.

Source document (simplified)

← USPTO Patent Grants

Unsupervised learning of object representations from video sequences using attention over space and time

Grant US12586353B2 Kind: B2 Mar 24, 2026

Assignee

GDM Holding LLC

Inventors

Rishabh Kabra, Daniel Zoran, Goker Erdogan, Antonia Phoebe Nina Creswell, Loic Matthey-de-l'Endroit, Matthew Botvinick, Alexander Lerchner, Christopher Paul Burgess

Abstract

A computer-implemented video generation neural network system, configured to determine a value for each of a set of object latent variables by sampling from a respective prior object latent distribution for the object latent variable. The system comprises a trained image frame decoder neural network configured to, for each pixel of each generated image frame and for each generated image frame time step process determined values of the object latent variables to determine parameters of a pixel distribution for each of the object latent variables, combine the pixel distributions for each of the object latent variables to determine a combined pixel distribution, and sample from the combined pixel distribution to determine a value for the pixel and for the time step.