Machine Learning

Error-Correcting Output Codes (ECOC)

About this project:

This project seeks to employ error-correcting output codes in machine learning to perform classification and incremental learning.

Goals

1. Find new efficient constructions of optimal ECOC matrices, including generalizations of the Hadamard construction and extensions to multi-symbol alphabets (e.g. ternary).

2. Derive sharp upper bounds on the ECOC error rate based on bit error rates of the ensemble of binary classifiers.

3. Find relevant data sets to experimentally verify the effectiveness of the ECOC approach.

Papers

  1. S. Ho, M. Marchiano, S. Zockoll, H. Nguyen, An Error-Correcting Output Code Framework for Lifelong Learning without a Teacher, ICTAI 2020, to appear.

CURRENT PROJECT MEMBERS:
Nicholas (Nico) Kaegi (undergraduate CS student, since March 2020)
Lucas Lavalva (undergraduate CS and Math student, since February 2019)
Dr. Shen-Shyang Ho (Department of Computer Science, since August 2018)
Dr. Hieu Nguyen (since August 2018)

PAST PROJECT MEMBERS
Mathew Marchiano (CS student, Aug 2018 - Dec 2019)
Scott Zockoll (CS student, Aug 2018 - Dec 2019)
Logan Borys (undergraduate Math student, May 2019 - June 2020)
Mohammed (Sarosh) Khan (undergraduate CS and Math student, Feb 2019 - Jan 2021)
Jonathan Moore (undergraduate CS student, since Sep 2020 - Feb 2021)

CONTACT INFORMATION:
Please email Hieu Nguyen (Rowan University) at nguyen@rowan.edu if you would like to learn more about this project.