Weighted maximum likelihood loss as a convenient shortcut to optimizing the F-measure of maximum entropy classifiers.
Georgi DimitroffLaura TolosiBorislav PopovGeorgi GeorgievPublished in: RANLP (2013)
Keyphrases
- maximum entropy
- maximum likelihood
- class conditional
- maximum entropy principle
- maximum entropy model
- markov models
- decision trees
- kernel logistic regression
- training set
- training data
- precision and recall
- em algorithm
- named entity recognition
- principle of maximum entropy
- maximum a posteriori
- iterative scaling
- support vector
- class labels
- conditional random fields
- expectation maximization
- gaussian distribution
- transformation based learning
- active learning
- minimum cross entropy
- naive bayes
- objective function
- natural language
- log likelihood
- likelihood function
- binary classifiers
- nearest neighbor
- roc curve
- hyperparameters