Torque-Based Structured Pruning for Deep Neural Networks

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published March 31st, 2026

Detected March 31st, 2026

Summary

USPTO granted Samsung Electronics Patent US12591778B2 for a torque-based structured pruning method for deep neural networks. The patent covers a machine learning training technique using torque-based constraints to concentrate weights in certain filters before pruning channels based on average weight calculations. The patent contains 26 claims and names three inventors.

View original document View source feed page

What changed

This is a USPTO patent grant for Samsung Electronics covering a method for torque-based structured pruning of deep neural networks. The patented approach applies torque-based constraints during ML model training to concentrate weights in specific filters before removing channels based on average weight calculations. Patent US12591778B2 (26 claims, Application No. 18052297, Filing Date November 3, 2022) was granted on March 31, 2026. CPC classifications include G06N 3/082, G06N 3/0464, G06N 20/00, G06N 20/10, and related neural network categories.

This is a patent grant notification rather than a regulatory requirement. Technology companies developing neural network systems or AI/ML products should consider reviewing the patent claims for potential freedom-to-operate implications. No compliance deadline or required actions exist for this document; it is informational in nature regarding Samsung's intellectual property rights.

Source document (simplified)

← USPTO Patent Grants

System and method for torque-based structured pruning for deep neural networks

Grant US12591778B2 Kind: B2 Mar 31, 2026

Assignee

Samsung Electronics Co., Ltd.

Inventors

Tien C. Bau, Arshita Gupta, Hrishikesh Deepak Garud

Abstract

A method includes accessing a machine learning model, the machine learning model trained using a torque-based constraint. The method also includes receiving an input from an input source and providing the input to the machine learning model. The method also includes receiving an output from the machine learning model. The method also includes instructing at least one action based on the output from the machine learning model. Training the machine learning model includes applying a torque-based constraint on one or more filters of the machine learning model, adjusting, based on applying the torque-based constraint, a first set of one or more filters of the machine learning model to have a higher concentration of weights than a second set of one or more filters of the machine learning model, and pruning at least one channel of the machine learning model based on an average weight for the at least one channel.