/Patent Tracker
NvidiaGranted patentAI & softwareInterest 4/10

Nvidia patents method to adjust neural network weights after removing parts

Nvidia describes a technique for simplifying artificial intelligence models by removing unnecessary components, a process known as pruning. The patent explains how the remaining parts of the network are mathematically adjusted to account for the removed sections. This ensures the simplified model continues to function correctly without needing to be rebuilt from scratch.

Why it mattersThis improves the efficiency of AI systems by allowing them to run faster and use less memory, which is useful for developers building software that needs to work on limited hardware.

In the filing’s own words

Apparatuses, systems, and techniques to prune neural network channels. In at least one embodiment, one or more data values computed as a result of executing a neural network are scaled according to one or more channels pruned from said neural network.

What exactly they claim (claim 1)

1. A processor, comprising: one or more circuits to train one or more neural networks by updating one or more weights of one or more neural networks based, at least in part, on a value representing a proportion of unpruned portions remaining after pruning one or more portions of the one or more neural networks.

Facts

Company
Nvidia (NVIDIA Corporation)
Official title
Scaling neural network weights
Patent number
US12737629B1
Application number
17842545
Filed
Jun 16, 2022
Granted
Sep 15, 2026
Inventors
4
Claims
35

Classified under

G06N 3/082 Physics & computingG06N 3/048 Physics & computing
Read the full filing (official PDF) ↗ Open in the Patent Tracker

More from Nvidia in AI & software

Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.