EasyScale: Elastic Training with Consistent Accuracy and Improved Utilization on GPUs.
Mingzhen LiWencong XiaoHailong YangBiao SunHanyu ZhaoShiru RenZhongzhi LuanXianyan JiaYi LiuYong LiWei LinDepei QianPublished in: SC (2023)
Keyphrases
- high accuracy
- computational cost
- prediction accuracy
- case study
- training phase
- training algorithm
- training speed
- neural network
- parallel architectures
- improved accuracy
- accuracy rate
- globally optimal
- training process
- improved algorithm
- high precision
- general purpose
- semi supervised
- classification accuracy
- training set
- feature selection
- information systems
- information retrieval