Patent · US Active

Nonlinear classification of data

US9224097B2 · kind B2 · utility

1Cited by

0References

19Claims

0Family size

Assignee

Meltwater Sweden AB · SE

Inventors

Babak Rasolzadeh · San Francisco, US
Oscar Danielsson · Stockholm, SE

Key dates

Filing date	Jun 5, 2012
Grant date	Dec 29, 2015
Priority date	—
Expiry date	Dec 21, 2032

Classification

Technology area (CPC G)Physics
CPC primaryG06V10/774
WIPO fieldComputer technology
WIPO sectorElectrical engineering

Abstract

The present invention relates to a method for nonlinear classification of high dimensional data by means of boosting, whereby a target class with significant intra-class variation is classified against a large background class, where the boosting algorithm produces a strong classifier, the strong classifier being a linear combination of weak classifiers. The present invention specifically teaches that weak classifiers classifiers h1, h2, that individually more often than not generate a positive on instances within the target class and a negative on instances outside of the target class, but that never generate a positive simultaneously on one and the same target instance, are categorized as a group of anti-correlated classifiers, and that the occurrence of anti-correlated classifiers from the same group will generate a negative.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.