Model-Generated Pretraining Signals Improves Zero-Shot Generalization of Text-to-Text Transformers.
Linyuan GongChenyan XiongXiaodong LiuPayal BajajYiqing XieAlvin CheungJianfeng GaoXia SongPublished in: CoRR (2023)
Keyphrases
- keywords
- formal model
- probabilistic model
- statistical model
- feature selection
- generation process
- text retrieval
- experimental data
- mathematical model
- computational model
- parameter estimation
- text mining
- probability distribution
- high level
- maximum likelihood
- theoretical framework
- management system
- objective function
- multiscale
- free text
- genetic algorithm