AStitch: enabling a new multi-dimensional optimization space for memory-intensive ML training and inference on modern SIMT architectures.
Zhen ZhengXuanda YangPengzhan ZhaoGuoping LongKai ZhuFeiwen ZhuWenyi ZhaoXiaoyong LiuJun YangJidong ZhaiShuaiwen Leon SongWei LinPublished in: ASPLOS (2022)
Keyphrases
- multi dimensional
- dimensional data
- maximum likelihood
- global optimization
- structured prediction
- training process
- optimization algorithm
- training set
- optimization problems
- training examples
- low dimensional
- space time
- evolutionary algorithm
- training algorithm
- test set
- memory management
- probabilistic inference
- memory requirements
- dual decomposition
- main memory
- optimization method
- sequential patterns
- data sets
- training samples
- search space
- bayesian networks
- training data
- neural network