NVIDIA Patent US12585275B1 for Neural Network Loss Scaling

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published March 24th, 2026

Detected March 24th, 2026

Summary

The USPTO has granted NVIDIA Corporation patent US12585275B1 for a method of scaling neural network loss values for object navigation. The patent, filed in 2019, details techniques for identifying features, fitting curves, and scaling loss based on real-world distance and curvature.

View original document View source feed page

What changed

The United States Patent and Trademark Office (USPTO) has granted patent US12585275B1 to NVIDIA Corporation, covering a novel method for scaling loss values in neural networks used for object navigation. The patent, which has a filing date of April 8, 2019, and an issue date of March 24, 2026, describes a system that obtains image data, identifies relevant features, and scales potential path loss values based on real-world distance and curvature. It also includes temporal smoothing and selection of paths with a high confidence value.

This patent grant is primarily an intellectual property development for NVIDIA and does not impose new regulatory obligations on other entities. However, it signifies a technological advancement in AI-driven navigation systems, potentially impacting the competitive landscape for companies developing autonomous systems, robotics, and advanced driver-assistance systems. Compliance officers in the technology sector should be aware of this patent as it relates to AI and machine learning innovations.

Source document (simplified)

← USPTO Patent Grants

Loss scaling for neural networks

Grant US12585275B1 Kind: B1 Mar 24, 2026

Assignee

NVIDIA Corporation

Inventors

Blythe Towal, Carolina Parada, Vijay Chintalapudi, Maroof Mohammed Farooq

Abstract

A navigation path can be determined for an object using one or more neural networks. In various embodiments, image data is obtained that is representative of an environment in which the object is to be navigated. Relevant features are identified from the image, and a curve fit to those features. Loss values for the potential paths are scaled based at least in part upon the distance of those features in the real world. This can include, in at least some embodiments, performing the scaling as a function of the curvature of the curve fit to the features. Temporal smoothing can be performed with respect to prior path predictions in order to prevent sudden changes in the predicted path. The paths are analyzed to select a path with a highest confidence value that also at least satisfies a minimum confidence criterion. The path can be converted into three-dimensional navigation information.