/Patent Tracker
NvidiaGranted patentAI & softwareInterest 3/10

Nvidia patents neural networks that map data dependencies in graphs

Nvidia describes a method where neural networks analyze directed graphs to create a "spacetime embedding" that maps how data points depend on one another. This process treats data connections like causal events in space and time, allowing the system to understand the direction and relationships between different pieces of information. The patent covers using this mapped structure to help other neural networks perform better calculations.

Why it mattersThis is a specialized technique for improving how AI models handle complex data structures, which could lead to more efficient processing in advanced computing tasks.

In the filing’s own words

Apparatuses, systems, and techniques to indicate data dependencies. In at least one embodiment, one or more neural networks are used to generate one or more indicators of one or more data dependencies and one or more indicators of direction of the one or more data dependencies.

What exactly they claim (claim 1)

1. A processor, comprising: one or more circuits to: receive data input for a directed graph; use one or more neural networks to generate a spacetime embedding of the directed graph, based on the data input for the directed graph, the spacetime embedding representing one or more data dependencies and direction of the one or more data dependencies within the directed graph as one or more causal connections between events in a spacetime, wherein to generate the spacetime embedding the one or more neural networks are to restrict a connected pair of nodes of the directed graph to be represented as causally-connected events within a same convex normal neighborhood of the spacetime such that the spacetime embedding is restricted to events connected on a time-directed curve within a threshold distance of one another in the same convex normal neighborhood of the spacetime; and perform inferencing using one or more neural networks and based at least on the spacetime embedding.

Facts

Company
Nvidia (NVIDIA Corporation)
Official title
Neural networks to indicate data dependencies
Patent number
US12769790B2
Application number
18098015
Filed
Jan 17, 2023
Granted
Oct 20, 2026
Inventors
2
Claims
20

Classified under

G06F 16/9024 Physics & computingG06N 3/09 Physics & computingG06N 3/044 Physics & computingG06N 3/0464 Physics & computingG06N 3/063 Physics & computing

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.