Learning multi-category classification in Bayesian framework

Atul Kanaujia, Dimitris Metaxas

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We propose an algorithm for Sparse Bayesian Classification for multi-class problems using Automatic Relevance Determination(ARD). Unlike other approaches which treat multiclass problem as multiple independent binary classification problem, we propose a method to learn the multiclass predictor directly. The usual approach of "one against rest" and "pairwise coupling" are not only computationally demanding during training stage but also generates dense classifiers which have greater tendency to overfit and have higher classification cost. In this paper we discuss the algorithmic implementation of Multiclass Classification model and compare it with other multi-class classifiers. We also empirically evaluate the classifier on viewpoint learning problem using features extracted from human silhouettes. Our experiments show that our algorithm generates sparser classifiers, with performance comparable to state-of-the-art multi-class classifier.

Original languageAmerican English
Title of host publicationComputer Vision - ACCV 2006 - 7th Asian Conference on Computer Vision, Proceedings
Pages255-264
Number of pages10
DOIs
StatePublished - 2006
Event7th Asian Conference on Computer Vision, ACCV 2006 - Hyderabad, India
Duration: Jan 13 2006Jan 16 2006

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume3851 LNCS

Other

Other7th Asian Conference on Computer Vision, ACCV 2006
Country/TerritoryIndia
CityHyderabad
Period1/13/061/16/06

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Learning multi-category classification in Bayesian framework'. Together they form a unique fingerprint.

Cite this