Hippie: A Data-Paralleled Pipeline Approach to Improve Memory-Efficiency and Scalability for Large DNN Training.
Xiangyu YeZhiquan LaiShengwei LiLei CaiDing SunLinbo QiaoDongsheng LiPublished in: ICPP (2021)
Keyphrases
- data sets
- data processing
- database
- data collection
- knowledge discovery
- data analysis
- synthetic data
- data structure
- data reduction
- data quality
- original data
- training samples
- raw data
- spatial data
- incoming data
- memory usage
- test data
- data distribution
- hidden markov models
- data streams
- statistical analysis
- databases
- training data
- input data
- data points