Amazon Patent for Data Shuffle Optimization

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published March 24th, 2026

Detected March 25th, 2026

Summary

The USPTO has granted Amazon Technologies, Inc. a patent (US12585961B1) for a data shuffle optimization algorithm designed to reduce operators in neural network models. This innovation aims to optimize neural network models by identifying and replacing specific data shuffle regions with more efficient affine shuffle operators.

View original document View source feed page

What changed

The United States Patent and Trademark Office (USPTO) has granted Amazon Technologies, Inc. patent US12585961B1 for an "Data shuffle optimization" algorithm. This patent covers a method for reducing data shuffle operators in neural network models by identifying single-entry single-exit (SESE) regions and replacing them with affine shuffle operators when implementation costs are lower. The patent abstract indicates this process optimizes neural network models by reducing the total number of operators in the data flow graph.

This patent grant is primarily of informational value to technology companies involved in AI and machine learning development. While it does not impose new regulatory obligations or compliance deadlines, it highlights a specific technological advancement by a major industry player. Companies developing or utilizing neural network models may find the disclosed optimization techniques relevant to their internal research and development efforts.

Source document (simplified)

← USPTO Patent Grants

Data shuffle optimization

Grant US12585961B1 Kind: B1 Mar 24, 2026

Assignee

Amazon Technologies, Inc.

Inventors

Hongbin Zheng, Jie Wang, Sheng Xu, Qingrui Liu, Pushkar Ratnalikar

Abstract

An optimization algorithm is disclosed to reduce the number of data shuffle operators in a data flow graph representing a neural network model for a neural network. The optimization algorithm can identify single-entry single-exit (SESE) regions in the data flow graph and select the SESE regions that comprise only the data shuffle operators. An affine map for each selected SESE region can be generated from an input of the selected SESE region to an output of the selected SESE region. An affine shuffle operator corresponding to an affine map for a selected SESE region can replace that SESE region if an implementation cost of the affine map is lower than the implementation cost of the SESE region. Thus, by replacing the selected SESE regions comprising multiple data shuffle operators with corresponding affine shuffle operators, it is possible to reduce the total number of operators in the data flow graph and optimize the neural network model.