Patent · US Active

Feature embeddings with relative locality for fast profiling of users on streaming data

US10685008B1 · kind B1 · utility

8Cited by
9References
36Claims
0Family size

Assignee

Inventors

Key dates

Filing dateMay 4, 2017
Grant dateJun 16, 2020
Priority date
Expiry dateAug 29, 2038

Classification

  • Technology area (CPC G)Physics
  • CPC primaryG10L17/18
  • WIPO fieldComputer technology
  • WIPO sectorElectrical engineering

Abstract

Systems and methods for classification using an explicit feature map or an approximate feature map based on a relative locality measure. In at least one embodiment, a method of authenticating a user operates data points having feature vectors pertaining to user events comprises selecting an approximate feature map based on a subset of features in each data point and a relative locality measure of a cluster including a plurality of the data points; mapping, to a feature space, the subset of features in each data point and a new data point pertaining to a phone call of the user using the selected approximate feature map; determining a classification of the new data point based on its relative locality measure with respect to a cluster in the feature space; storing the classification of the new data point in a memory device; and authenticating the user during the phone call.

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