The patent badge is an abbreviated version of the USPTO patent document. The patent badge does contain a link to the full patent document.

The patent badge is an abbreviated version of the USPTO patent document. The patent badge covers the following: Patent number, Date patent was issued, Date patent was filed, Title of the patent, Applicant, Inventor, Assignee, Attorney firm, Primary examiner, Assistant examiner, CPCs, and Abstract. The patent badge does contain a link to the full patent document (in Adobe Acrobat format, aka pdf). To download or print any patent click here.

Date of Patent:
Jun. 02, 2026

Filed:

May. 03, 2023
Applicant:

Microsoft Technology Licensing, Llc, Redmond, WA (US);

Inventors:

Fereshte Khani, San Francisco, CA (US);

Zexue He, La Jolla, CA (US);

Marco Tulio Correia Ribeiro, Seattle, WA (US);

Assignee:
Attorneys:
Primary Examiner:
Int. Cl.
CPC ...
G06F 40/40 (2020.01); G06N 5/022 (2023.01);
U.S. Cl.
CPC ...
G06F 40/40 (2020.01); G06N 5/022 (2013.01);
Abstract

A system and method and for method for optimizing performance of a natural language processing (NLP) model includes clustering a validation dataset used in training the NLP model into a plurality of clusters; measuring a generalization in context parameter for one or more of the plurality of clusters; measuring an interference in context parameter for one or more of the plurality of clusters; and identifying a cluster, from among the plurality of clusters, for data augmentation, based on the measured generalization in context parameter and the measured interference in context parameter. Once a cluster is identified, a prompt is generated for submission as an input to a large language model (LLM) to prompt the LLM to automatically generate synthetic training data for the identified cluster, before the prompt is provided to the LLM and synthetic training data is received from the LLM. The synthetic training data is then labeled by a human before being used to further train the NLP model to improve the performance of the NLP model with respect to the identified cluster.


Find Patent Forward Citations

Loading…