Hi-Speed DNN Training with Espresso: Unleashing the Full Potential of Gradient Compression with Near-Optimal Usage Strategies.
Zhuang WangHaibin LinYibo ZhuT. S. Eugene NgPublished in: EuroSys (2023)
Keyphrases
- training process
- data compression
- high speed
- training speed
- compression scheme
- neural network
- arithmetic coding
- compression rate
- usage patterns
- edge detection
- supervised learning
- multiresolution
- training set
- real time
- training examples
- compression ratio
- image compression
- training algorithm
- hidden markov models
- gradient information