NvidiaGranted patentChips & hardwareInterest 3/10

Nvidia patents chip method to run image math as tensor contractions

Nvidia describes a processor design that handles image analysis by converting standard convolution steps into tensor contractions. The system creates higher-dimensional data structures from the image and performs specific mathematical operations on them to produce the final output.

Why it mattersThis is a technical optimization for how chips process visual data, which may slightly improve efficiency for developers building image recognition software.

In the filing’s own words

Apparatuses, systems, and techniques to convert between tensor convolution and tensor contraction operations. In at least one embodiment, one or more convolution operations are performed on image data by at least contracting one or more tensors to generate one or more feature maps.

What exactly they claim (claim 1)

1. One or more processors, comprising: circuitry to generate a feature map corresponding to image data, wherein to generate the feature map, the circuitry is to at least: generate one or more second tensors from one or more first tensors generated from the image data, wherein the one or more second tensors have a greater number of dimensions than the one or more first tensors; and simulate, by at least performing a contraction of the one or more second tensors, performance of one or more convolutional operations using the one or more first tensors.

Facts

Company
Nvidia (NVIDIA Corporation)
Official title
Processor and system to convert tensor operations in machine learning
Patent number
US12792902B2
Application number
16559544
Filed
Sep 3, 2019
Granted
Nov 10, 2026
Inventors
2
Claims
32

Classified under

G06N 3/08 Physics & computingG06F 7/57 Physics & computing

More from Nvidia in Chips & hardware

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.