Gradient Ascent Post-training Enhances Language Model Generalization.
Dongkeun YoonJoel JangSungdong KimMinjoon SeoPublished in: ACL (2) (2023)
Keyphrases
- language model
- gradient ascent
- language modeling
- probabilistic model
- n gram
- document retrieval
- information retrieval
- smoothing methods
- cross entropy
- retrieval model
- expectation maximization
- mixture model
- training set
- test collection
- query expansion
- unsupervised learning
- translation model
- training algorithm
- supervised learning
- web search engines
- bayesian inference
- exponential family
- information retrieval systems