Dynamic quantization for energy efficient deep learning

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published March 31st, 2026

Detected March 31st, 2026

Summary

The USPTO granted patent US12591771B2 to Qualcomm Incorporated on March 31, 2026, covering dynamic quantization methods for energy-efficient deep learning inference. The patent, invented by Randy Ardywibowo, Venkata Ravi Kiran Dayana, and Hau Hwang, describes a technique for quantizing neural network layer parameters based on input content during inference. The patent contains 28 claims and was filed as Application No. 17488261.

View original document View source feed page

What changed

Qualcomm has been granted US Patent 12,591,771 B2 for a method of dynamic quantization in deep neural networks. The invention enables energy-efficient inference by quantizing network parameters based on the content of layer inputs during the inference stage, rather than using fixed quantization schemes. The patent covers DNN layers that receive content-associated inputs and perform tasks using quantized parameters.

Patent grants are informational IP rights documents that do not impose compliance obligations on third parties. Technology companies developing deep learning or AI inference systems may consider this patent when assessing freedom-to-operate for quantization techniques. No regulatory action is required.

Source document (simplified)

← USPTO Patent Grants

Dynamic quantization for energy efficient deep learning

Grant US12591771B2 Kind: B2 Mar 31, 2026

Assignee

QUALCOMM Incorporated

Inventors

Randy Ardywibowo, Venkata Ravi Kiran Dayana, Hau Hwang

Abstract

A method performed by a deep neural network (DNN) includes receiving, at a layer of the DNN during an inference stage, a layer input comprising content associated with a DNN input received at the DNN. The method also includes quantizing one or more parameters of a plurality of parameters associated with the layer based on the content of the layer input. The method further includes performing a task corresponding to the DNN input, the task performed with the one or more one quantized parameters.