Linear Classifiers are Nearly Optimal When Hidden Variables Have Diverse Effect.
Nader H. BshoutyPhilip M. LongPublished in: COLT (2009)
Keyphrases
- hidden variables
- linear classifiers
- bayesian networks
- probabilistic model
- em algorithm
- multi class
- generative model
- latent variables
- bayesian inference
- hyperplane
- generalization error
- missing values
- principal components
- worst case
- closed form
- lower bound
- model selection
- higher order
- upper bound
- pattern recognition
- data sets