Meta patents method to trim training data for machine learning models
Meta describes a system that identifies data samples that are too far from a model's decision boundary and removes them from the training set. This creates a smaller, modified dataset used to train the machine learning model.
In the filing’s own words
Systems, apparatuses and methods provide technology that identifies a plurality of data samples of a data set. In response to a trigger, the technology identifies a first subset of samples of the plurality of data samples that are outside a range associated with a decision boundary of a machine learning model, and removes the first subset of samples from the data set to generate a modified training set. The technology train the machine learning model based on the modified training set.
What exactly they claim (claim 1)
1. At least one computer readable storage medium comprising a set of instructions, which when executed by a computing device, cause the computing device to: identify a plurality of data samples of a data set; in response to a trigger, identify a first subset of samples of the plurality of data samples that are outside a range associated with a decision boundary of a machine learning model; categorize a first sample of the first subset of samples as being a confident sample based on a label of the first sample and a classification of the first sample; determine that the first sample is outside the range associated with the decision boundary; group the first sample into the first subset based at least in part on the first sample being categorized as a confident sample that is determined to be outside the range associated with the decision boundary; and remove the first subset of samples from the data set to generate a modified training set; and train the machine learning model based on the modified training set.
Facts
- Company
- Meta (Meta Platforms, Inc.)
- Official title
- Ranked pruning of data set to train machine learning models
- Patent number
- US12743660B2
- Application number
- 18169462
- Filed
- Feb 15, 2023
- Granted
- Sep 22, 2026
- Inventors
- 5
- Claims
- 17
Classified under
G06N 20/00 Physics & computingMore from Meta in AI & software
- Meta patents AI summaries for notifications on glasses and watches Sep 29, 2026
- Meta patents avatar editing that keeps your face while changing the rest Sep 10, 2026
- Meta patents group video effects that change when you interact with them Oct 20, 2026
- Meta patents search method that uses learning objectives to find content Oct 1, 2026
- Meta patents method to track message order in large AI clusters Sep 1, 2026
Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.