EFLOPS: Algorithm and System Co-Design for a High Performance Distributed Training Platform.
Jianbo DongZheng CaoTao ZhangJianxi YeShaochuang WangFei FengLi ZhaoXiaoyong LiuLiuyihan SongLiwei PengYiqun GuoXiaowei JiangLingbo TangYin DuYingya ZhangPan PanYuan XiePublished in: HPCA (2020)
Keyphrases
- improved algorithm
- dynamic programming
- learning algorithm
- preprocessing
- input data
- detection algorithm
- optimal solution
- computational complexity
- training phase
- objective function
- experimental evaluation
- computational cost
- high accuracy
- simulated annealing
- high efficiency
- optimization algorithm
- search algorithm
- tree structure
- training algorithm
- computationally efficient
- worst case
- probabilistic model
- significant improvement
- distributed architecture
- times faster
- training examples
- bit rate
- np hard
- reinforcement learning