DecentLaM: Decentralized Momentum SGD for Large-batch Deep Training.
Kun YuanYiming ChenXinmeng HuangYingya ZhangPan PanYinghui XuWotao YinPublished in: CoRR (2021)
Keyphrases
- stochastic gradient descent
- learning rate
- training speed
- deep architectures
- peer to peer
- training samples
- batch mode
- training process
- supervised learning
- online learning
- test set
- artificial neural networks
- training set
- training algorithm
- online algorithms
- cooperative
- multi agent
- reinforcement learning
- text classification
- genetic algorithm
- convergence speed
- learning algorithm