Local-to-global learning for iterative training of production SLU models on new features.
Yulia GrishinaDaniil SorokinPublished in: NAACL-HLT (Industry Papers) (2022)
Keyphrases
- learned models
- structured prediction
- prior knowledge
- learning models
- supervised learning
- online learning
- discriminatively trained
- deep architectures
- predictive power
- classification models
- online training
- feature representations
- learning process
- learning tasks
- co occurrence
- skill acquisition
- low level
- learning problems
- learning algorithm
- hidden variables
- training phase
- learning systems
- semi supervised learning
- image features
- learning speed
- training program
- reinforcement learning