Description:

AUTOMATIC LABELER ASSIGNMENT

INV-1216

INVENTORS: Jennifer G. DY

Link to PDF

Description

Currently, several machine based algorithms are used to learn concepts in the presence of simultaneous labels. However, such algorithms are limited by their similar reliability of labelers across the data points. This approach provides a novel method for modeling data comprising multiple data points processed by multiple labelers in order to produce labeler error/accuracy estimates along with building a data classifier for the new data points.

 

Value Proposition

The method:

•Saves a significant amount of time and money to label new data points

•Determines the level of expertise/appropriateness of labelers for new data points

•Builds a mathematical/probabilistic model to predict the label of a data point

•Can be implemented as software embedded onto a programmed storage device such as RAM, ROM, DVD, CD, etc.

 

Intellectual Property status

Utility Application 12/893,008

License status 

Available for license

Patent Information:
For Information, Contact:
Mark Saulich
Senior Commercialization Manager
Northeastern University
m.saulich@neu.edu
Inventors:
Jennifer Dy
Glenn Fung
Romer Rosales
Yan Yan
Keywords:
Data Mining
Information Processing
Information Technology
Internet
Multimedia
Software
Software and Algorithms
WWW