SRU-Q: Hardware-friendly Stochastic Rounding Unit-based Gradient Quantization for CNN Training.
Sangbeom JeongDahun ChoiHyun KimPublished in: AICAS (2024)
Keyphrases
- control unit
- low cost
- hardware and software
- cellular neural networks
- supervised learning
- monte carlo
- real time
- processing units
- training set
- approximation algorithms
- computer systems
- training examples
- neural network
- training phase
- training samples
- edge detection
- online learning
- linear programming
- mixed integer
- computational complexity
- multiscale
- linear programming relaxation
- computer software
- vlsi implementation
- e learning
- friendly interface