Mitigating harm in language models with conditional-likelihood filtration.
Helen NgoCooper RaterinkJoão G. M. AraújoIvan ZhangCarol ChenAdrien MorisotNicholas FrosstPublished in: CoRR (2021)
Keyphrases
- language model
- conditional likelihood
- probabilistic model
- markov networks
- posterior probability
- conditional random fields
- bayesian network classifiers
- language modeling
- logistic regression
- saddle point
- n gram
- information retrieval
- query expansion
- graphical models
- retrieval model
- discriminative learning
- bayesian networks
- naive bayes
- mixture model
- graphical structure
- generative model
- class labels
- maximum likelihood
- classification error
- probability distribution
- smoothing methods
- training data
- machine learning