ICMLA
2007
14 years 9 days ago
2007
Given any generative classifier based on an inexact density model, we can define a discriminative counterpart that reduces its asymptotic error rate, while increasing the estima...