Machine Learning Distribution Anomaly Detection Program Patent

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published March 31st, 2026

Detected March 31st, 2026

Summary

USPTO granted Patent US12591808B2 to Fujitsu Limited on March 31, 2026. The patent covers a machine learning detection program that identifies distribution differences between datasets by comparing model outputs against loss function gradients. The patent contains 18 claims and is classified under CPC G06N 20/00.

View original document View source feed page

What changed

USPTO granted Patent US12591808B2 to Fujitsu Limited on March 31, 2026. The patent covers a detection program using machine learning to detect differences between data distributions by inputting data into a second ML model and comparing the output against a threshold based on loss function gradients. The invention involves generating a second ML model from first data and first results, inputting second data to acquire second results, and detecting distribution differences based on the comparison. The patent has 18 claims and was filed on March 22, 2023.

This is a patent grant notice that does not impose new compliance obligations on regulated entities. Companies developing machine learning-based anomaly detection systems or data distribution analysis tools should consider potential freedom-to-operate implications. No immediate regulatory action is required, but R&D and product development teams may wish to assess whether their technologies could implicate this patent.

Source document (simplified)

← USPTO Patent Grants

Computer-readable recording medium storing detection program, detection method, and detection device

Grant US12591808B2 Kind: B2 Mar 31, 2026

Assignee

Fujitsu Limited

Inventors

Hiroaki Kingetsu

Abstract

A non-transitory computer-readable recording medium stores a detection program for causing a computer to execute processing including: inputting a plurality of pieces of second data into a second machine learning model generated by machine learning based on a plurality of pieces of first data and a first result output from a first machine learning model according to an input of the plurality of pieces of first data; acquiring a second result output from the second machine learning model according to the input of the plurality of pieces of second data; and detecting a difference between a distribution of the plurality of pieces of first data and a distribution of the plurality of pieces of second data, based on comparison between a value calculated based on the second result and a gradient of a loss function of the second machine learning model with a threshold.