ActNN: Reducing Training Memory Footprint via 2-Bit Activation Compressed Training.
Jianfei ChenLianmin ZhengZhewei YaoDequan WangIon StoicaMichael W. MahoneyJoseph GonzalezPublished in: ICML (2021)
Keyphrases
- memory footprint
- training algorithm
- training process
- training set
- training samples
- training phase
- training examples
- neural network
- data compression
- learning environment
- training data
- case study
- genetic algorithm
- data mining
- data structure
- active learning
- supervised learning
- information processing
- test set
- information systems
- learning algorithm
- computer software
- real time