Name
Title | ||
|---|---|---|
A Model Selection Criterion for Classification: Application to HMM Topology Optimization |
Abstract | ||
|---|---|---|
This paper proposes a model selection criterion for classificationproblems. The criterion focuses on selecting modelsthat are discriminant instead of models based on the Occam'srazor principle of parsimony between accurate modelingand complexity. The criterion, dubbed DiscriminativeInformation Criterion (DIC), is applied to the optimizationof Hidden Markov Model topology aimed at the recognitionof cursively-handwritten digits. The results show that DIC-generatedmodels achieve 18% relative improvement in per-formancefrom a baseline system generated by the BayesianInformation Criterion (BIC). |
Year | DOI | Venue |
|---|---|---|
2003 | 10.1109/ICDAR.2003.1227641 | ICDAR-1 |
Keywords | Field | DocType |
Bayes methods,handwritten character recognition,hidden Markov models,image classification,optimisation,Bayesian information criterion,HMM topology optimization,Occam razor principle,classification problems,cursively-handwritten digit recognition,discriminative information criterion,hidden Markov model topology,model selection criterion | Bayesian information criterion,Pattern recognition,Computer science,Markov model,Model selection,occam,Topology optimization,Artificial intelligence,Contextual image classification,Hidden Markov model,Discriminative model,Machine learning | Conference |
ISSN | ISBN | Citations |
1520-5363 | 0-7695-1960-1 | 27 |
PageRank | References | Authors |
1.63 | 5 | 1 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Alain Biem | 1 | 288 | 18.64 |