DAPPLE: a pipelined data parallel approach for training large models.
Shiqing FanYi RongChen MengZongyan CaoSiyu WangZhen ZhengChuan WuGuoping LongJun YangLixue XiaLansong DiaoXiaoyong LiuWei LinPublished in: PPoPP (2021)
Keyphrases
- data sets
- data collection
- computer systems
- synthetic data
- raw data
- data analysis
- xml documents
- statistical methods
- end users
- data mining techniques
- training samples
- accurate models
- training dataset
- incomplete data
- experimental data
- missing data
- training examples
- knowledge discovery
- prior knowledge
- high quality
- databases
- statistical analysis
- data processing
- data distribution
- missing values
- data structure
- neural network model
- historical data
- decision trees
- empirical data