Linear classifiers are nearly optimal when hidden variables have diverse effects.
Nader H. BshoutyPhilip M. LongPublished in: Mach. Learn. (2012)
Keyphrases
- hidden variables
- linear classifiers
- probabilistic model
- em algorithm
- hyperplane
- bayesian inference
- multi class
- generative model
- bayesian networks
- latent variables
- worst case
- generalization error
- hyperparameters
- probability distribution
- reinforcement learning
- data mining
- nearest neighbor
- kernel function
- semi supervised
- learning process
- data analysis
- posterior distribution
- feature selection
- computer vision