Smart training and smart deployment of machine learning models

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published April 7th, 2026

Detected April 7th, 2026

Summary

The USPTO granted patent US12596940B2 to GE Precision Healthcare LLC covering systems and techniques for smart training and deployment of machine learning models. The invention computes feature distributions of training data candidates and identifies a strict subset whose distributions match the full dataset, enabling more efficient ML model training. The patent includes 20 claims under CPC classifications G06N 5/041, G06N 20/00, G06N 3/08, G06N 3/04, and G06F 17/18.

View original document View source feed page

What changed

The USPTO issued patent US12596940B2 to GE Precision Healthcare LLC for systems facilitating smart training of machine learning models. The invention identifies a subset of training data whose feature distributions match those of the complete training dataset, reducing computational requirements while maintaining model quality. The system computes feature distributions, selects a strict subset with matching statistical properties, and trains ML models on this optimized subset.

Companies developing machine learning technologies should review this patent to assess potential licensing needs or competitive implications. The patent's focus on data subset selection for efficient ML training may affect approaches to machine learning model development, particularly in healthcare and medical technology applications where GE Precision Healthcare operates.

What to do next

Monitor for updates on related patent applications in ML training
Review patent claims for potential licensing or design-around opportunities

Source document (simplified)

← USPTO Patent Grants

Smart training and smart deployment of machine learning models

Grant US12596940B2 Kind: B2 Apr 07, 2026

Assignee

GE Precision Healthcare LLC

Inventors

Sidharth Abrol, Aanchal Mongia, Abhijit Patil

Abstract

Systems/techniques that facilitate smart training and smart deployment of machine learning models are provided. In various embodiments, a system can access a first set of data candidates that are available for training of a machine learning model. In various aspects, the system can compute at least one feature distribution of the first set of data candidates. In various instances, the system can identify, in the first set of data candidates, a strict subset of data candidates, wherein at least one feature distribution of the strict subset of data candidates matches the at least one feature distribution of the first set of data candidates. In various cases, the system can train the machine learning model on the strict subset of data candidates.